Skip to content

C
chatbot canifa
Commit 94484304 authored by Vũ Hoàng Anh's avatar Vũ Hoàng Anh
Browse files
Options

update

parent f9b10ebe
Show whitespace changes
Inline Side-by-side
Showing with 3684 additions and 4090 deletions
backend/agent/__init__.py
View file @ 94484304
......@@ -2,9 +2,14 @@
Fashion Q&A Agent Package
"""
from .graph import build_graph
from .models import AgentConfig, AgentState, get_config
def build_graph(*args, **kwargs):
from .graph import build_graph as _build_graph
return _build_graph(*args, **kwargs)
__all__ = [
"AgentConfig",
"AgentState",
......
backend/agent/controller.py
View file @ 94484304
"""
Fashion Q&A Agent Controller
Langfuse will auto-trace via LangChain integration (no code changes needed).
Fashion Q&A Agent Controller (Refactored for 2-Agent Architecture)
Uses CANIFAGraph (Classifier -> Stylist) for high-performance deterministic routing.
"""
import asyncio
import json
import logging
import re
import time
import uuid
import time
from typing import List, Dict, Any
from fastapi import BackgroundTasks
from langchain_core.messages import AIMessage, HumanMessage
from langchain_core.runnables import RunnableConfig
from langfuse import Langfuse, get_client as get_langfuse
from common.cache import redis_cache
......@@ -21,18 +19,16 @@ from common.langfuse_client import async_flush_langfuse, get_callback_handler
from config import DEFAULT_MODEL, REDIS_CACHE_TURN_ON
from .controller_helpers import (
extract_and_save_user_insight,
finalize_user_insight_after_stream,
check_response_cache,
save_user_insight_to_redis,
load_user_insight_from_redis,
)
from .graph import build_graph
from .helper import extract_product_ids, handle_post_chat_async
from .models import AgentState, get_config
from .streaming_callback import ProductIDStreamingCallback
from .graph import get_graph_manager
from .helper import handle_post_chat_async
from .models import get_config
logger = logging.getLogger(__name__)
async def chat_controller(
query: str,
identity_key: str,
......@@ -40,328 +36,113 @@ async def chat_controller(
model_name: str = DEFAULT_MODEL,
images: list[str] | None = None,
return_user_insight: bool = False,
is_authenticated: bool = False, # New: để track user vs guest
device_id: str | None = None, # Device ID từ request (có cả khi authenticated)
skip_cache: bool = False, # Skip response cache (for dev endpoint)
is_authenticated: bool = False,
device_id: str | None = None,
skip_cache: bool = False,
) -> dict:
"""
Controller main logic for non-streaming chat requests.
Flow:
1. Check cache (if enabled) → HIT: return cached response
2. MISS: Call LLM → Save to cache → Return response
3. user_insight extract:
- return_user_insight=False (prod): background task (non-blocking)
- return_user_insight=True (dev): extract ngay và return luôn
Args:
identity_key: Key for saving/loading history (identity.history_key)
Guest: device_id, User: user_id
return_user_insight: If True, extract and return user_insight immediately (dev mode)
is_authenticated: True if user logged in, False if guest
Controller logic using the new 2-Agent Architecture.
"""
logger.info("chat_controller start: model=%s, key=%s", model_name, identity_key)
logger.info("🚀 [ChatController] Starting Turn | Identity: %s", identity_key)
# ====================== CACHE LAYER ======================
# 1. Cache Layer
if REDIS_CACHE_TURN_ON and not skip_cache:
cached_response = await redis_cache.get_response(user_id=identity_key, query=query)
if cached_response:
logger.info("CACHE HIT key=%s", identity_key)
cached_insight = None
if identity_key:
cached_insight = await load_user_insight_from_redis(identity_key)
if cached_insight:
cached_response = {**cached_response, "user_insight": cached_insight}
# --- Langfuse: log cache hit trace ---
try:
langfuse = get_langfuse()
if langfuse:
trace_id = Langfuse.create_trace_id()
session_id = f"{identity_key}-cache"
tags = ["chatbot", "cache-hit", "user:authenticated" if is_authenticated else "user:anonymous"]
langfuse.trace(
id=trace_id,
name="chat-request-cache-hit",
user_id=identity_key,
session_id=session_id,
tags=tags,
input={"query": query, "user_id": identity_key},
output={
"ai_response": (cached_response.get("ai_response", "") or "")[:500],
"product_count": len(cached_response.get("product_ids", [])),
"source": "cache",
},
metadata={
"device_id": device_id,
"customer_id": identity_key if is_authenticated else None,
"model": model_name,
"cache_hit": True,
},
cached_res = await check_response_cache(
identity_key, query, background_tasks, model_name, is_authenticated, device_id
)
except Exception:
logger.debug("Failed to log cache hit trace", exc_info=True)
memory = await get_conversation_manager()
background_tasks.add_task(
handle_post_chat_async,
memory=memory,
identity_key=identity_key,
human_query=query,
ai_response=cached_response,
)
return {**cached_response, "cached": True}
if cached_res:
return cached_res
# 2. Setup Config & Memory
config = get_config()
config.model_name = model_name
graph = build_graph(config)
manager = get_graph_manager(config)
memory = await get_conversation_manager()
history_dicts = await memory.get_chat_history(identity_key, limit=15, include_product_ids=False)
messages = [
HumanMessage(content=m["message"]) if m["is_human"] else AIMessage(content=m["message"]) for m in history_dicts
# 3. Load History & Insight
history_dicts = await memory.get_chat_history(identity_key, limit=10, include_product_ids=False)
history = [
HumanMessage(content=m["message"]) if m["is_human"] else AIMessage(content=m["message"])
for m in history_dicts
][::-1]
# ====================== GRAPH EXECUTION ======================
user_insight = None
if identity_key:
user_insight = await load_user_insight_from_redis(identity_key)
initial_state: AgentState = {
"user_query": HumanMessage(content=query),
"history": messages,
"user_id": identity_key,
"user_insight": user_insight,
"images_embedding": [],
"ai_response": None,
}
run_id = str(uuid.uuid4())
# 4. Tracing Setup (Langfuse)
trace_id = Langfuse.create_trace_id()
session_id = f"{identity_key}-{run_id[:8]}"
tags = ["chatbot", "user:authenticated" if is_authenticated else "user:anonymous"]
streaming_callback = ProductIDStreamingCallback()
start_time = time.time()
event_count = 0
ai_text_response = ""
final_product_ids = []
all_accumulated_messages = []
seen_message_ids = set()
async def consume_events():
nonlocal event_count, all_accumulated_messages, seen_message_ids
async for event in graph.astream(initial_state, config=exec_config):
event_count += 1
for _node, node_state in event.items():
if isinstance(node_state, dict) and "messages" in node_state:
for msg in node_state["messages"]:
msg_id = id(msg)
if msg_id not in seen_message_ids:
seen_message_ids.add(msg_id)
all_accumulated_messages.append(msg)
early_response = False
# --- Langfuse: wrap everything in a named trace with custom trace_id ---
langfuse = get_langfuse()
observation_ctx = (
langfuse.start_as_current_observation(
as_type="span",
name="chat-request",
trace_context={"trace_id": trace_id},
)
if langfuse
else None
)
if observation_ctx:
span = observation_ctx.__enter__()
span.update_trace(
name="chat-request",
user_id=identity_key,
session_id=session_id,
tags=tags,
input={"query": query, "user_id": identity_key},
metadata={
"device_id": device_id,
"customer_id": identity_key if is_authenticated else None,
"model": model_name,
},
)
session_id = f"{identity_key}-{str(uuid.uuid4())[:8]}"
tags = ["stylist_pro", "user:authenticated" if is_authenticated else "user:anonymous"]
# CallbackHandler MUST be created INSIDE the observation context
# so it inherits the trace_id and nests under "chat-request"
langfuse_handler = get_callback_handler()
exec_config = RunnableConfig(
configurable={
exec_config = {
"configurable": {
"user_id": identity_key,
"device_id": device_id,
"model_name": model_name,
"transient_images": images or [],
"run_id": run_id,
"trace_id": trace_id,
},
run_id=run_id,
metadata={
"callbacks": [langfuse_handler] if langfuse_handler else [],
"metadata": {
"trace_id": trace_id,
"session_id": session_id,
"device_id": device_id,
"customer_id": identity_key if is_authenticated else None,
},
callbacks=[langfuse_handler, streaming_callback] if langfuse_handler else [streaming_callback],
)
}
}
# 5. Graph Invocation (Non-Streaming High-Performance)
# We use .chat() method which includes enrichment and insight management
try:
stream_task = asyncio.create_task(consume_events())
start_time = time.time()
if return_user_insight:
await stream_task
ai_text_response = streaming_callback.ai_response_text
final_product_ids = streaming_callback.product_skus
else:
wait_task = asyncio.create_task(streaming_callback.product_found_event.wait())
done, pending = await asyncio.wait(
[stream_task, wait_task],
return_when=asyncio.FIRST_COMPLETED,
timeout=30.0,
# We wrap the call to trace it manually in Langfuse if needed,
# but the LangChain callbacks in graph nodes handle most of it.
result = await manager.chat(
user_message=query,
user_insight=user_insight,
history=history,
config=exec_config
)
ai_text_response = streaming_callback.ai_response_text
final_product_ids = streaming_callback.product_skus
if streaming_callback.product_ids_found:
early_response = True
if wait_task in pending:
wait_task.cancel()
finally:
# Update trace output before closing
if observation_ctx:
try:
elapsed = time.time() - start_time
span.update_trace(
output={
"ai_response": (ai_text_response or "")[:500],
"product_count": len(final_product_ids) if final_product_ids else 0,
"elapsed_s": round(elapsed, 2),
"early_return": early_response,
},
)
except Exception:
pass
try:
observation_ctx.__exit__(None, None, None)
except Exception:
pass
# Flush Langfuse to ensure traces are sent immediately
try:
await async_flush_langfuse()
except Exception:
pass
try:
# Ensure stream completes for tool messages
if not early_response and not stream_task.done():
try:
await stream_task
except asyncio.CancelledError:
logger.warning("stream_task cancelled")
# Extract ai_response from streaming content (fallbacks)
if early_response and not ai_text_response:
raw_content = streaming_callback.accumulated_content
# Strip Codex reasoning objects before parsing
raw_content = ProductIDStreamingCallback.strip_reasoning(raw_content)
if raw_content:
try:
raw_normalized = raw_content.replace("{{", "{").replace("}}", "}")
ai_json = json.loads(raw_normalized)
if isinstance(ai_json, dict):
ai_text_response = ai_json.get("ai_response", ai_text_response)
if not final_product_ids and isinstance(ai_json.get("product_ids"), list):
final_product_ids = [str(s) for s in ai_json["product_ids"]]
except json.JSONDecodeError:
ai_match = re.search(r'"ai_response"\s*:\s*"((?:[^"\\]|\\.)*)"', raw_content)
if ai_match:
ai_text_response = ai_match.group(1).replace('\\"', '"').replace("\\n", "\n")
if not ai_text_response and all_accumulated_messages:
for msg in reversed(all_accumulated_messages):
if isinstance(msg, AIMessage) and msg.content:
# Responses API may return content as list
content = msg.content
if isinstance(content, list):
content = "".join(str(c.get("text", c) if isinstance(c, dict) else c) for c in content)
# Strip Codex reasoning objects
content = ProductIDStreamingCallback.strip_reasoning(content)
ai_text_response = content
break
# Parse JSON-wrapped ai_response
# Ensure ai_text_response is str (Responses API may return list)
if isinstance(ai_text_response, list):
ai_text_response = "".join(str(c.get("text", c) if isinstance(c, dict) else c) for c in ai_text_response)
# Strip Codex reasoning objects before JSON parse
ai_text_response = ProductIDStreamingCallback.strip_reasoning(ai_text_response)
if ai_text_response and ai_text_response.lstrip().startswith("{"):
try:
ai_normalized = ai_text_response.replace("{{", "{").replace("}}", "}")
ai_json = json.loads(ai_normalized)
if isinstance(ai_json, dict):
ai_text_response = ai_json.get("ai_response", ai_text_response)
if not final_product_ids and isinstance(ai_json.get("product_ids"), list):
final_product_ids = [str(s) for s in ai_json["product_ids"]]
except json.JSONDecodeError:
# Regex fallback for Codex {{/}} braces that break JSON parse
ai_match = re.search(r'"ai_response"\s*:\s*"((?:[^"\\]|\\.)*)"\s*,\s*"product_ids"', ai_text_response, re.DOTALL)
if ai_match:
ai_text_response = ai_match.group(1).replace('\\"', '"').replace("\\n", "\n")
pid_match = re.search(r'"product_ids"\s*:\s*\[(.*?)\]', ai_text_response if not ai_match else ai_normalized, re.DOTALL)
if pid_match and not final_product_ids:
final_product_ids = re.findall(r'"([^"]+)"', pid_match.group(1))
# Extract & filter products
enriched_products = []
if all_accumulated_messages:
all_extracted = extract_product_ids(all_accumulated_messages)
if final_product_ids:
mentioned = set(final_product_ids)
enriched_products = [p for p in all_extracted if p.get("sku") in mentioned]
else:
enriched_products = all_extracted
logger.info("✅ Graph Finished | Time: %.2fs | Products: %d", elapsed, len(result.get("products", [])))
except Exception as e:
logger.error("Error extracting products: %s", e)
enriched_products = []
logger.error("❌ Graph execution failed: %s", e, exc_info=True)
raise e
user_insight_dict = None
if return_user_insight:
content = streaming_callback.accumulated_content
if content and identity_key:
user_insight_dict = await extract_and_save_user_insight(content, identity_key)
elif background_tasks:
background_tasks.add_task(
finalize_user_insight_after_stream,
stream_task=stream_task,
streaming_callback=streaming_callback,
identity_key=identity_key,
trace_id=trace_id,
)
# 6. Response Preparation
ai_response = result.get("ai_response", "")
final_products = result.get("products", [])
updated_insight = result.get("updated_insight")
response_payload = {
"ai_response": ai_text_response,
"product_ids": enriched_products,
"ai_response": ai_response,
"product_ids": final_products, # Full objects as expected by modern UI
"user_insight": updated_insight,
"lead_stage": result.get("lead_stage"),
"elapsed_ms": result.get("elapsed_ms")
}
if user_insight_dict is not None:
response_payload["user_insight"] = user_insight_dict
# 7. Post-Chat Tasks (Async)
# Save Insight to Redis
if updated_insight:
import json
insight_str = json.dumps(updated_insight, ensure_ascii=False)
background_tasks.add_task(save_user_insight_to_redis, identity_key, insight_str)
# Save to Cache
if REDIS_CACHE_TURN_ON and not skip_cache:
await redis_cache.set_response(user_id=identity_key, query=query, response_data=response_payload, ttl=300)
background_tasks.add_task(
redis_cache.set_response,
user_id=identity_key,
query=query,
response_data=response_payload,
ttl=300
)
# Save to History
background_tasks.add_task(
handle_post_chat_async,
memory=memory,
......@@ -369,4 +150,8 @@ async def chat_controller(
human_query=query,
ai_response=response_payload,
)
# Flush Langfuse
background_tasks.add_task(async_flush_langfuse)
return {**response_payload, "cached": False}
backend/agent/controller_helpers.py
View file @ 94484304
......@@ -7,8 +7,14 @@ import json
import logging
import re
import time
from fastapi import BackgroundTasks
from langfuse import Langfuse, get_client as get_langfuse
from common.cache import redis_cache
from common.conversation_manager import get_conversation_manager
from common.langfuse_client import async_flush_langfuse, get_callback_handler
from agent.helper import handle_post_chat_async, extract_product_ids
logger = logging.getLogger(__name__)
......@@ -150,8 +156,6 @@ def extract_products_from_state(final_state_snapshot: dict) -> list[dict]:
Returns:
list[dict]: List of product objects with name, price, sale_price, etc.
"""
from .helper import extract_product_ids
enriched_products = []
try:
......@@ -176,3 +180,69 @@ def extract_products_from_state(final_state_snapshot: dict) -> list[dict]:
logger.error(f"❌ Error extracting products from state: {e}")
return enriched_products
async def check_response_cache(
identity_key: str,
query: str,
background_tasks: BackgroundTasks,
model_name: str,
is_authenticated: bool,
device_id: str | None = None,
) -> dict | None:
"""
Helper to handle the cache layer logic: Redis lookup, Insight merge, Langfuse tracing, and background tasks.
Returns a dict response if HIT, otherwise None.
"""
# 1. Check Redis for cached response
cached_response = await redis_cache.get_response(user_id=identity_key, query=query)
if not cached_response:
return None
logger.info(f"🚀 [Cache Layer] HIT | key={identity_key}")
# 2. Merge User Insight (if any)
cached_insight = await load_user_insight_from_redis(identity_key)
if cached_insight:
cached_response = {**cached_response, "user_insight": cached_insight}
# 3. Log Langfuse Trace (Cache Hit)
try:
langfuse = get_langfuse()
if langfuse:
trace_id = Langfuse.create_trace_id()
session_id = f"{identity_key}-cache"
tags = ["chatbot", "cache-hit", "user:authenticated" if is_authenticated else "user:anonymous"]
langfuse.trace(
id=trace_id,
name="chat-request-cache-hit",
user_id=identity_key,
session_id=session_id,
tags=tags,
input={"query": query, "user_id": identity_key},
output={
"ai_response": (cached_response.get("ai_response", "") or "")[:500],
"product_count": len(cached_response.get("product_ids", [])),
"source": "cache",
},
metadata={
"device_id": device_id,
"customer_id": identity_key if is_authenticated else None,
"model": model_name,
"cache_hit": True,
},
)
except Exception:
logger.debug("Failed to log cache hit trace", exc_info=True)
# 4. Background Task (Persist conversation history for cache hit)
memory = await get_conversation_manager()
background_tasks.add_task(
handle_post_chat_async,
memory=memory,
identity_key=identity_key,
human_query=query,
ai_response=cached_response,
)
return {**cached_response, "cached": True}
backend/agent/graph.py
View file @ 94484304
"""
Fashion Q&A Agent Graph
LangGraph workflow với clean architecture.
Tất cả resources (LLM, Tools) khởi tạo trong __init__.
Sử dụng ConversationManager (Postgres) để lưu history thay vì checkpoint.
"""
import hashlib
import logging
from datetime import datetime
from typing import Any
import json
import time
from typing import Dict, Any, List
from langchain_core.language_models import BaseChatModel
from langchain_core.messages import HumanMessage
from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
from langgraph.graph import StateGraph, END
from langchain_core.messages import HumanMessage, AIMessage, BaseMessage
from langchain_core.runnables import RunnableConfig
from langgraph.cache.memory import InMemoryCache
from langgraph.graph import END, StateGraph
from langgraph.prebuilt import ToolNode
from langgraph.types import CachePolicy
from common.llm_factory import create_llm
# Modular Imports from the refined nodes
from agent.models import AgentConfig, get_config
from agent.nodes.models_node import StylistProState
from agent.nodes.classifier_node import classifier_node
from agent.nodes.tools_node import tools_node
from agent.nodes.stylist_node import stylist_node
from agent.nodes.utils import _extract_text
from .models import AgentConfig, AgentState, get_config
from .prompt_utils import get_system_prompt_template
from .tools.get_tools import get_all_tools
# StarRocks connection for Enrichment
from common.starrocks_connection import get_db_connection
logger = logging.getLogger(__name__)
# Constants for Product Enrichment (Hardcoded for clarity and decoupling)
TABLE_NAME = "shared_source.magento_product_dimension_with_text_embedding"
SELECT_COLUMNS = """
magento_ref_code, internal_ref_code, product_name,
sale_price, original_price, product_image_url_thumbnail,
product_web_url, size_scale, gender_by_product, product_line_vn
"""
def route_after_classifier(state: StylistProState) -> str:
"""Điều hướng dựa trên early_exit."""
if state.get("early_exit"):
return END
return "stylist"
class CANIFAGraph:
"""
Fashion Q&A Agent Graph Manager.
"""
def __init__(
self,
config: AgentConfig | None = None,
llm: BaseChatModel | None = None,
tools: list | None = None,
):
self.config = config or get_config()
self._compiled_graph: Any | None = None
self.llm: BaseChatModel = llm or create_llm(
model_name=self.config.model_name, api_key=self.config.openai_api_key, streaming=True
)
self.all_tools = tools or get_all_tools()
self.llm_with_tools = self.llm.bind_tools(self.all_tools, strict=True)
self.cache = InMemoryCache()
# Chain caching: avoid rebuilding ChatPromptTemplate every turn
self._cached_chain = None
self._cached_prompt_hash: str | None = None
Fashion Q&A Agent Graph (Version 2.0).
Architecture: Classifier (Lightweight) -> Stylist (Heavyweight) -> END
def _build_chain(self, system_prompt_template: str):
"""Build chain with dynamic system prompt (fetched from Langfuse per request).
Caches the chain and only rebuilds when prompt content changes.
All logic is modularized into nodes for maximum performance and readability.
"""
# Check if prompt changed via hash comparison
prompt_hash = hashlib.md5(system_prompt_template.encode()).hexdigest()
if self._cached_chain is not None and self._cached_prompt_hash == prompt_hash:
return self._cached_chain
logger.info("🔄 Building new chain (prompt hash changed)" if self._cached_chain else "🔧 Building initial chain")
prompt_template = ChatPromptTemplate.from_messages(
[
("system", system_prompt_template),
(
"system",
"===== USER INSIGHT (TỪ TURN TRƯỚC) =====\n⚡ BẮT BUỘC: Đọc [LAST_ACTION] bên dưới để hiểu context turn trước, rồi tự suy ra bước tiếp theo!\n\n{user_insight}\n=====================================",
),
("system", "Chat History:"),
MessagesPlaceholder(variable_name="history"),
("system", "Current Query:"),
MessagesPlaceholder(variable_name="user_query"),
MessagesPlaceholder(variable_name="messages"),
]
)
self._cached_chain = prompt_template | self.llm_with_tools
self._cached_prompt_hash = prompt_hash
return self._cached_chain
async def _agent_node(self, state: AgentState, config: RunnableConfig) -> dict:
"""Agent node - Chỉ việc đổ dữ liệu riêng vào khuôn đã có sẵn."""
messages = state.get("messages", [])
history = state.get("history", [])
user_query = state.get("user_query")
transient_images = config.get("configurable", {}).get("transient_images", [])
if transient_images and user_query:
# Inject images into user_query as multimodal content
# GPT-4o / GPT-5 vision: text + image_url blocks
text_content = user_query.content if isinstance(user_query.content, str) else str(user_query.content)
# Append hint so LLM knows images are attached
text_content += "\n\n[📸 Có ảnh sản phẩm kèm theo - Hãy mô tả CHI TIẾT sản phẩm trong ảnh (loại, màu, kiểu dáng, hình in, chất liệu) rồi gọi data_retrieval_tool để tìm sản phẩm tương tự.]"
multimodal_content = [{"type": "text", "text": text_content}]
for img in transient_images:
# Support both base64 and URL
if img.startswith("data:"):
image_url = img
elif img.startswith("http"):
image_url = img
else:
image_url = f"data:image/jpeg;base64,{img}"
multimodal_content.append({
"type": "image_url",
"image_url": {"url": image_url, "detail": "auto"}
})
logger.info(f"📸 [IMAGE] Image size: {len(img)} chars")
user_query = HumanMessage(content=multimodal_content)
logger.info(f"📸 [IMAGE] Injected {len(transient_images)} image(s) into user_query")
# Invoke chain with history, user_query, messages (scratchpad), and user_insight
user_insight_text = (
state.get("user_insight")
or "⚠️ TRẠNG THÁI KHỞI TẠO: Chưa có User Insight từ lịch sử. Hãy bắt đầu thu thập thông tin mới (Nếu thiếu thông tin thì ghi 'Chưa rõ')."
)
# Fetch prompt from Langfuse (cached 5min by SDK) + build chain (cached by hash)
current_date_str = datetime.now().strftime("%d/%m/%Y")
system_prompt_template = get_system_prompt_template()
chain = self._build_chain(system_prompt_template)
response = await chain.ainvoke(
{
"date_str": current_date_str,
"history": history,
"user_query": [user_query] if user_query else [],
"messages": messages,
"user_insight": user_insight_text,
}
)
return {"messages": [response], "ai_response": response}
def _should_continue(self, state: AgentState) -> str:
"""Routing: tool nodes hoặc end."""
last_message = state["messages"][-1]
if not hasattr(last_message, "tool_calls") or not last_message.tool_calls:
logger.info("🏁 Agent finished")
return "end"
def __init__(self, config: AgentConfig | None = None):
self.config = config or get_config()
self._compiled = None
tool_names = [tc["name"] for tc in last_message.tool_calls]
logger.info(f"🔄 → tools: {tool_names}")
return "tools"
def build(self):
if self._compiled:
return self._compiled
def build(self) -> Any:
"""Build và compile LangGraph workflow."""
if self._compiled_graph is not None:
return self._compiled_graph
workflow = StateGraph(AgentState)
workflow = StateGraph(StylistProState)
# Nodes
workflow.add_node("agent", self._agent_node)
workflow.add_node("tools", ToolNode(self.all_tools))
workflow.add_node("classifier", classifier_node)
workflow.add_node("tools", tools_node) # New dedicated tools node
workflow.add_node("stylist", stylist_node)
# Entry point
workflow.set_entry_point("classifier")
# Routing logic
def route_after_classifier(state: StylistProState):
if state.get("early_exit"):
return END
if state.get("tool_name_used"):
return "tools"
return "stylist"
# Edges
workflow.set_entry_point("agent")
workflow.add_conditional_edges(
"agent",
self._should_continue,
{"tools": "tools", "end": END},
"classifier",
route_after_classifier,
{
"tools": "tools",
"stylist": "stylist",
END: END
}
)
workflow.add_edge("tools", "agent")
self._compiled_graph = workflow.compile(cache=self.cache) # No Checkpointer
logger.info("✅ Graph compiled (Langfuse callback will be per-run)")
workflow.add_edge("tools", "stylist") # After tools, always go to stylist
workflow.add_edge("stylist", END)
return self._compiled_graph
self._compiled = workflow.compile()
logger.info("✅ CANIFAGraph compiled (3-Agent Architecture: Classifier -> Tools -> Stylist)")
return self._compiled
@property
def graph(self) -> Any:
return self.build()
async def chat(
self,
user_message: str,
user_insight: str | None = None,
history: List[BaseMessage] | None = None,
config: RunnableConfig | None = None,
) -> Dict[str, Any]:
"""
Main entry point with StarRocks Product Enrichment.
Ensures all products mentioned by AI have full metadata (image, price, url).
"""
start_time = time.time()
graph = self.build()
messages = history.copy() if history else []
messages.append(HumanMessage(content=user_message))
# --- Singleton & Public API ---
_instance: list[CANIFAGraph | None] = [None]
initial_state = {
"messages": messages,
"user_insight": user_insight,
"tool_result": None,
"tool_name_used": None,
"updated_insight": None,
"lead_stage": None,
"stage_injection": None,
"product_ids": [],
"early_exit": False,
"diagnostics": [],
}
# Invoke Graph
result = await graph.ainvoke(initial_state, config=config)
elapsed_ms = (time.time() - start_time) * 1000
# Extract AI Response
ai_response = ""
for msg in reversed(result["messages"]):
if isinstance(msg, AIMessage):
ai_response = _extract_text(msg.content)
break
# --- PRODUCT ENRICHMENT LAYER ---
final_products = []
ai_product_ids = result.get("product_ids", [])
tool_result_raw = result.get("tool_result")
product_dict = {}
# 1. Map products already in tool result (if any)
if tool_result_raw:
try:
parsed = json.loads(tool_result_raw)
if parsed.get("status") == "success":
all_products = parsed.get("products", []) or parsed.get("results", [])
for p in all_products:
sku = str(p.get("magento_ref_code") or p.get("sku") or p.get("sku_code") or "").upper().strip()
if sku:
if "sku" not in p: p["sku"] = sku
product_dict[sku] = p
# Also map outfit recommendations
recs = p.get("outfit_recommendations", [])
if isinstance(recs, list):
for r in recs:
r_sku = str(r.get("match_product_code") or r.get("sku") or "").upper().strip()
if r_sku:
if "sku" not in r: r["sku"] = r_sku
product_dict[r_sku] = r
except Exception: pass
# 2. Enrich missing or incomplete product data via StarRocks
wanted_ids = [pid.upper().strip() for pid in ai_product_ids if pid]
missing_ids = [pid for pid in wanted_ids if pid not in product_dict or not product_dict[pid].get("image")]
if missing_ids:
try:
db = get_db_connection()
skus = list(dict.fromkeys(missing_ids))
base_skus = list(dict.fromkeys([s.split("-")[0] for s in skus]))
keys = list(dict.fromkeys(skus + base_skus))
placeholders = ",".join(["%s"] * len(keys))
sql = f"""
SELECT {SELECT_COLUMNS}
FROM {TABLE_NAME}
WHERE UPPER(magento_ref_code) IN ({placeholders})
OR UPPER(internal_ref_code) IN ({placeholders})
LIMIT 200
"""
rows = await db.execute_query_async(sql, params=tuple(keys + keys))
for r in rows or []:
card = {
"sku": (r.get("magento_ref_code") or r.get("internal_ref_code") or "").strip(),
"name": r.get("product_name", ""),
"price": int(r.get("sale_price") or 0),
"original_price": int(r.get("original_price") or 0),
"image": r.get("product_image_url_thumbnail", ""),
"url": r.get("product_web_url", ""),
"sizes": r.get("size_scale", ""),
"gender": r.get("gender_by_product", ""),
"product_line": r.get("product_line_vn", ""),
}
product_dict[card["sku"].upper()] = card
except Exception as e:
logger.warning(f"⚠️ StarRocks Enrichment failed: {e}")
# 3. Assemble final products in the order requested by AI
seen = set()
for pid in wanted_ids:
if pid in product_dict and pid not in seen:
final_products.append(product_dict[pid])
seen.add(pid)
return {
"response": ai_response,
"ai_response": ai_response, # compatibility
"elapsed_ms": round(elapsed_ms, 2),
"lead_stage": result.get("lead_stage"),
"updated_insight": result.get("updated_insight"),
"user_insight": result.get("updated_insight"), # compatibility
"product_ids": ai_product_ids,
"products": final_products,
"pipeline": result.get("diagnostics", [])
}
def build_graph(config: AgentConfig | None = None, llm: BaseChatModel | None = None, tools: list | None = None) -> Any:
"""Get compiled graph (Singleton usage)."""
# Use singleton to avoid rebuilding graph on every request
manager = get_graph_manager(config, llm, tools)
return manager.build()
# --- Singleton & Compatibility API ---
_instance: List[CANIFAGraph | None] = [None]
def reset_chain_cache():
"""Clear the singleton instance to force a rebuild."""
global _instance
_instance[0] = None
logger.info("🔄 Graph cache reset")
def get_graph_manager(
config: AgentConfig | None = None, llm: BaseChatModel | None = None, tools: list | None = None
) -> CANIFAGraph:
"""Get CANIFAGraph instance (Auto-rebuild if model config changes).
def reset_graph():
"""Alias for reset_chain_cache for compatibility."""
reset_chain_cache()
Prompt is now fetched dynamically per request from Langfuse,
so no need to rebuild graph when prompt changes.
"""
# 1. New Instance if Empty
def build_graph(config: AgentConfig | None = None) -> Any:
"""Compatibility function for controller."""
global _instance
if _instance[0] is None:
_instance[0] = CANIFAGraph(config, llm, tools)
logger.info(f"✨ Graph Created: {_instance[0].config.model_name} (prompts from Langfuse)")
return _instance[0]
# 2. Check for Model Config Changes only
is_model_changed = config and config.model_name != _instance[0].config.model_name
if is_model_changed:
logger.info(f"🔄 Rebuilding Graph: Model ({_instance[0].config.model_name}->{config.model_name})")
_instance[0] = CANIFAGraph(config, llm, tools)
return _instance[0]
_instance[0] = CANIFAGraph(config)
return _instance[0].build()
def get_graph_manager(config: AgentConfig | None = None) -> CANIFAGraph:
global _instance
if _instance[0] is None:
_instance[0] = CANIFAGraph(config)
return _instance[0]
def reset_graph() -> None:
"""Reset singleton for testing."""
_instance[0] = None
def reset_chain_cache() -> None:
"""Reset only the cached chain (when prompt changes).
Keeps the graph/LLM/tools intact, only forces chain rebuild on next request.
"""
if _instance[0] is not None:
_instance[0]._cached_chain = None
_instance[0]._cached_prompt_hash = None
logger.info("🔄 Chain cache cleared — will rebuild on next request")
backend/agent/models.py
View file @ 94484304
......@@ -2,7 +2,7 @@ from typing import Annotated, Any, TypedDict
from langchain_core.messages import BaseMessage
from langgraph.graph.message import add_messages
from pydantic import BaseModel
from pydantic import BaseModel, Field
import config as global_config
......@@ -56,3 +56,71 @@ def get_config() -> AgentConfig:
langfuse_secret_key=global_config.LANGFUSE_SECRET_KEY,
langfuse_base_url=global_config.LANGFUSE_BASE_URL,
)
# ==============================================================================
# STYLIST PRO MODELS (Architecture 2.0)
# ==============================================================================
class RawSearchArgs(BaseModel):
"""Lane 1: Trích xuất từ khóa gốc, giữ nguyên ý định khách, sửa lỗi typo."""
raw_text: str = Field(default="", description="Core keywords from user message")
class StructuredSearchArgs(BaseModel):
"""Lane 2: Suy luận bộ lọc có cấu trúc từ ngữ cảnh."""
product_line_vn: list[str] = Field(default_factory=list, description="Áo phông, Quần jean...")
gender_target: str | None = Field(default=None, description="men/women/boy/girl/unisex")
age_group: str | None = Field(default=None, description="adult/kid")
master_color: str | None = Field(default=None, description="Màu sắc suy luận")
tags: list[str] = Field(default_factory=list)
keywords: list[str] = Field(default_factory=list)
price_min: int | None = Field(default=None)
price_max: int | None = Field(default=None)
discount_min: int | None = Field(default=None)
discovery_mode: str | None = Field(default=None)
class ClassifierSearchArgs(BaseModel):
"""Tổng hợp 2 Lane Search."""
literal: RawSearchArgs = Field(default_factory=RawSearchArgs)
inferred: StructuredSearchArgs = Field(default_factory=StructuredSearchArgs)
magento_ref_code: str | None = Field(default=None)
class StylistProInsight(BaseModel):
"""Bộ nhớ khách hàng - 12 trường cốt lõi."""
USER: str | None = Field(default="Chưa rõ")
TARGET: str | None = Field(default="Chưa rõ")
GOAL: str | None = Field(default="Chưa rõ")
CONSTRAINTS: str | None = Field(default="")
STAGE: str | None = Field(default="BROWSE")
STAGE_NUM: int = Field(default=1)
TONE: str | None = Field(default="Friendly")
BEHAVIORAL_HINTS: list[str] = Field(default_factory=list)
LATEST_PRODUCT_INTEREST: str | None = Field(default="")
LAST_ACTION: str | None = Field(default="")
SUMMARY_HISTORY: str | None = Field(default="")
class ClassifierOutput(BaseModel):
"""Kết quả từ node Classifier."""
reasoning: str
tool_name: str | None = None
search_args: ClassifierSearchArgs | None = None
tool_args: dict[str, Any] | None = None
ai_response: str | None = None
product_ids: list[str] = Field(default_factory=list)
class StylistOutput(BaseModel):
"""Kết quả từ node Stylist."""
ai_response: str
product_ids: list[str] = Field(default_factory=list)
user_insight: StylistProInsight
class StylistProState(TypedDict):
"""Trạng thái luồng chạy của Stylist Pro."""
messages: Annotated[list[BaseMessage], add_messages]
user_insight: str | None
tool_name_used: str | None
tool_result: str | None
updated_insight: dict[str, Any] | None
lead_stage: dict[str, Any] | None
stage_injection: str | None
product_ids: list[str]
early_exit: bool | None
diagnostics: Annotated[list[dict[str, Any]], lambda x, y: x + y]
backend/agent/nodes/__init__.py
View file @ 94484304
"""
Agent Nodes Package
Agent Nodes Package (Architecture 2.0)
"""
from .agent import agent_node
from .classifier_node import classifier_node
from .stylist_node import stylist_node
from .models_node import StylistProState
__all__ = ["agent_node"]
__all__ = [
"classifier_node",
"stylist_node",
"StylistProState"
]
backend/agent/nodes/classifier_node.py 0 → 100644
View file @ 94484304
import json
import logging
import time
from langchain_core.messages import AIMessage, HumanMessage, SystemMessage
from langchain_core.runnables import RunnableConfig
from common.llm_factory import create_llm
from agent.nodes.models_node import StylistProState, ClassifierOutput
from agent.nodes.utils import _extract_text, _format_history, parse_llm_json
from agent.prompt_module.stylist_pro_prompts import STYLIST_PRO_CLASSIFIER_PROMPT
logger = logging.getLogger(__name__)
async def classifier_node(state: StylistProState, config: RunnableConfig):
"""
Classifier Node: Định tuyến intent và thực thi Tool inline.
Sử dụng data_retrieval_tool làm công cụ tìm kiếm chính.
"""
llm = create_llm(model_name=config.get("configurable", {}).get("model_name"), streaming=False)
messages = state["messages"]
user_query = _extract_text(messages[-1].content)
user_insight = state.get("user_insight") or "{}"
history_text = _format_history(messages[:-1])
# Context preparation
products_context = ""
try:
if isinstance(user_insight, str) and user_insight.startswith("{"):
_ins = json.loads(user_insight)
latest = _ins.get("LATEST_PRODUCT_INTEREST", "")
summary = _ins.get("SUMMARY_HISTORY", "")
if latest: products_context += f"\n=== SP ĐÃ GIỚI THIỆU ===\n{latest}\n"
if summary: products_context += f"\n=== TÓM TẮT HỘI THOẠI ===\n{summary}\n"
except Exception: pass
context_block = (
f"=== TIN NHẮN HIỆN TẠI ===\n{user_query}\n\n"
f"=== USER INSIGHT ===\n{user_insight}\n\n"
f"{products_context}"
f"=== LỊCH SỬ ===\n{history_text}"
)
prompt = [
SystemMessage(content=STYLIST_PRO_CLASSIFIER_PROMPT),
HumanMessage(content=context_block),
]
start_time = time.time()
response = await llm.ainvoke(prompt, config=config)
elapsed = (time.time() - start_time) * 1000
data = parse_llm_json(response.content)
# --- PHÒNG THỦ: Dọn dẹp dữ liệu trước khi nạp vào Pydantic ---
if "search_args" in data and "inferred" in data["search_args"]:
inferred = data["search_args"]["inferred"]
# Các trường mong đợi String/Int nhưng hay bị AI trả về List rỗng []
str_fields = ["gender_target", "age_group", "master_color", "price_min", "price_max"]
for field in str_fields:
if field in inferred and (isinstance(inferred[field], list) or isinstance(inferred[field], dict)):
if isinstance(inferred[field], list):
inferred[field] = inferred[field][0] if len(inferred[field]) > 0 else None
else:
inferred[field] = None
# Xử lý product_line_vn (đảm bảo luôn là List)
if "product_line_vn" in inferred and isinstance(inferred["product_line_vn"], str):
inferred["product_line_vn"] = [inferred["product_line_vn"]]
output = ClassifierOutput(**data)
# Tool Registry
tool_name = output.tool_name
tool_result = None
diagnostics = [{
"step": "classifier",
"label": f"🎯 Classifier -> {tool_name or 'Early Exit'}",
"content": output.reasoning,
"elapsed_ms": round(elapsed)
}]
# Prepare tool_args
tool_args = {}
if tool_name:
# Case A: Search Tools (Data Retrieval) - UPDATED FOR DUAL-LANE
if (tool_name == "lead_search_tool" or tool_name == "data_retrieval_tool") and output.search_args:
inf = output.search_args.inferred
tool_args = {
"searches": [
{
"literal": {"raw_text": output.search_args.literal.raw_text},
"inferred": {
"product_line_vn": inf.product_line_vn or [],
"gender_by_product": inf.gender_target,
"age_by_product": inf.age_group,
"master_color": inf.master_color,
"price_min": inf.price_min,
"price_max": inf.price_max,
"discount_min": 0,
"discount_max": 100,
"discovery_mode": None
}
}
]
}
# Case B: Store Search (Ensure 'location' exists)
elif tool_name == "canifa_store_search":
loc = (output.tool_args or {}).get("location") or user_query
tool_args = {"location": loc}
# Case C: Knowledge Search (Ensure 'query' exists)
elif tool_name == "canifa_knowledge_search":
q = (output.tool_args or {}).get("query") or user_query
tool_args = {"query": q}
# Case D: Other Specific Tools
elif output.tool_args:
tool_args = output.tool_args
return {
"tool_name_used": tool_name,
"tool_args": tool_args,
"early_exit": tool_name is None and output.ai_response is not None,
"product_ids": output.product_ids,
"messages": [AIMessage(content=output.ai_response)] if output.ai_response else [],
"diagnostics": diagnostics
}
backend/agent/nodes/models_node.py 0 → 100644
View file @ 94484304
from typing import Annotated, Any, TypedDict, Optional, List
from pydantic import BaseModel, Field
from langchain_core.messages import BaseMessage
from langgraph.graph.message import add_messages
# ==============================================================================
# SEARCH MODELS
# ==============================================================================
class LiteralSearchArgs(BaseModel):
raw_text: str = ""
class InferredSearchArgs(BaseModel):
product_line_vn: List[str] = Field(default_factory=list)
gender_target: Optional[str] = None
age_group: Optional[str] = None
master_color: Optional[str] = None
tags: List[str] = Field(default_factory=list)
keywords: List[str] = Field(default_factory=list)
price_min: Optional[int] = None
price_max: Optional[int] = None
class ClassifierSearchArgs(BaseModel):
literal: LiteralSearchArgs = Field(default_factory=LiteralSearchArgs)
inferred: InferredSearchArgs = Field(default_factory=InferredSearchArgs)
magento_ref_code: Optional[str] = None
# ==============================================================================
# NODE OUTPUT MODELS
# ==============================================================================
class ClassifierOutput(BaseModel):
reasoning: str
tool_name: Optional[str] = None
search_args: Optional[ClassifierSearchArgs] = None
tool_args: Optional[dict] = None
ai_response: Optional[str] = None
product_ids: List[str] = Field(default_factory=list)
class StylistProInsight(BaseModel):
USER: Optional[str] = "Chưa rõ"
TARGET: Optional[str] = "Chưa rõ"
GOAL: Optional[str] = "Chưa rõ"
CONSTRAINTS: Optional[str] = None
STAGE: Optional[str] = "BROWSE"
STAGE_NUM: int = 1
TONE: Optional[str] = "Friendly"
LATEST_PRODUCT_INTEREST: Optional[str] = ""
SUMMARY_HISTORY: Optional[str] = ""
BEHAVIORAL_HINTS: List[str] = Field(default_factory=list)
class StylistOutput(BaseModel):
ai_response: str
product_ids: List[str] = Field(default_factory=list)
user_insight: StylistProInsight
# ==============================================================================
# STATE DEFINITION
# ==============================================================================
class StylistProState(TypedDict):
messages: Annotated[list[BaseMessage], add_messages]
user_insight: Optional[str]
tool_name_used: Optional[str]
tool_args: Optional[dict] # Added for 3-node flow
tool_result: Optional[str]
updated_insight: Optional[dict]
lead_stage: Optional[dict]
stage_injection: Optional[str]
product_ids: List[str]
early_exit: Optional[bool]
diagnostics: Annotated[list[dict[str, Any]], lambda x, y: x + y]
backend/agent/nodes/stylist_node.py 0 → 100644
View file @ 94484304
import json
import logging
import time
from typing import Dict, Any
from langchain_core.messages import AIMessage, HumanMessage, SystemMessage
from langchain_core.runnables import RunnableConfig
from common.llm_factory import create_llm
from agent.nodes.models_node import StylistProState, StylistOutput
from agent.nodes.utils import _extract_skus_from_text, _dedupe_preserve_order, _format_tool_result, parse_llm_json
from agent.prompt_module.stylist_pro_prompts import STYLIST_PRO_RESPONDER_PROMPT, format_stylist_pro_injection
logger = logging.getLogger(__name__)
async def stylist_node(state: StylistProState, config: RunnableConfig):
"""
Stylist Node: Sinh câu trả lời tư vấn và cập nhật Insight khách hàng.
"""
llm = create_llm(model_name=config.get("configurable", {}).get("model_name"), streaming=False)
insight_dict = json.loads(state.get("user_insight") or "{}")
injection = format_stylist_pro_injection(insight_dict)
system_content = STYLIST_PRO_RESPONDER_PROMPT.format(stage_injection=injection)
messages = [SystemMessage(content=system_content)]
# History context (3 turns)
human_count = 0
recent = []
for msg in reversed(state["messages"][:-1]):
if isinstance(msg, HumanMessage):
recent.insert(0, msg)
human_count += 1
elif isinstance(msg, AIMessage) and not (hasattr(msg, "tool_calls") and msg.tool_calls):
recent.insert(0, msg)
if human_count >= 3: break
messages.extend(recent)
messages.append(state["messages"][-1])
# Tool context
tool_context = _format_tool_result(state.get("tool_result"))
if tool_context:
messages.append(HumanMessage(content=f"=== KẾT QUẢ TRA CỨU ===\n{tool_context}"))
# Extra instructions
messages.append(HumanMessage(content="REMINDER: Output ONLY JSON. List ALL SKUs in product_ids. Ensure Top+Bottom for combos."))
start_time = time.time()
response = await llm.ainvoke(messages, config=config)
elapsed = (time.time() - start_time) * 1000
data = parse_llm_json(response.content)
output = StylistOutput(**data)
# Enrichment & Deduplication
text_skus = _extract_skus_from_text(output.ai_response)
final_product_ids = _dedupe_preserve_order((output.product_ids or []) + text_skus)
# Insight mapping
insight_raw = output.user_insight
insight_dict = insight_raw.model_dump() if hasattr(insight_raw, 'model_dump') else insight_raw
lead_stage = {
"stage": insight_dict.get("STAGE_NUM", 1),
"stage_name": insight_dict.get("STAGE", "BROWSE"),
"tone_directive": insight_dict.get("TONE", "Friendly"),
"behavioral_hints": insight_dict.get("BEHAVIORAL_HINTS", []),
}
diag = {
"step": "stylist",
"label": "💬 Stylist Response",
"content": output.ai_response[:200],
"elapsed_ms": round(elapsed)
}
return {
"updated_insight": insight_dict,
"lead_stage": lead_stage,
"stage_injection": format_stylist_pro_injection(insight_dict),
"product_ids": final_product_ids,
"messages": [AIMessage(content=output.ai_response)],
"diagnostics": [diag]
}
backend/agent/nodes/tools_node.py 0 → 100644
View file @ 94484304
import json
import logging
import time
from langchain_core.runnables import RunnableConfig
from agent.nodes.models_node import StylistProState
from agent.tools.data_retrieval_tool import data_retrieval_tool
logger = logging.getLogger(__name__)
def _get_tool_registry():
from agent.tools.brand_knowledge_tool import canifa_knowledge_search
from agent.tools.check_is_stock import check_is_stock
from agent.tools.promotion_canifa_tool import canifa_get_promotions
from agent.tools.store_search_tool import canifa_store_search
return {
"lead_search_tool": data_retrieval_tool,
"data_retrieval_tool": data_retrieval_tool,
"canifa_knowledge_search": canifa_knowledge_search,
"canifa_get_promotions": canifa_get_promotions,
"canifa_store_search": canifa_store_search,
"check_is_stock": check_is_stock,
}
async def tools_node(state: StylistProState, config: RunnableConfig):
"""
Tools Node: Thực thi Tool dựa trên tool_name_used và tool_args từ Classifier.
"""
tool_name = state.get("tool_name_used")
tool_args = state.get("tool_args") or {}
tool_result = None
diagnostics = []
if not tool_name:
return {"tool_result": None, "diagnostics": []}
tool_registry = _get_tool_registry()
tool_fn = tool_registry.get(tool_name)
if tool_fn:
t_start = time.time()
try:
logger.info(f"🛠️ [ToolsNode] Executing '{tool_name}' with args: {tool_args}")
# Execute Tool
tool_result = await tool_fn.ainvoke(tool_args)
tool_elapsed = (time.time() - t_start) * 1000
diagnostics.append({
"step": "tool_result",
"label": f"📦 {tool_name}",
"content": f"Success: {str(tool_result)[:100]}...",
"raw_json": tool_result,
"elapsed_ms": round(tool_elapsed)
})
logger.info(f"✅ [ToolsNode] '{tool_name}' finished in {round(tool_elapsed)}ms")
except Exception as e:
logger.error(f"❌ [ToolsNode] Execution Error in '{tool_name}': {e}")
tool_result = json.dumps({"status": "error", "message": str(e)})
diagnostics.append({
"step": "tool_error",
"label": f"❌ {tool_name} Error",
"content": str(e),
"elapsed_ms": 0
})
else:
logger.warning(f"⚠️ [ToolsNode] Unknown tool: '{tool_name}'")
return {
"tool_result": tool_result,
"diagnostics": diagnostics
}
backend/agent/nodes/utils.py 0 → 100644
View file @ 94484304
import json
import logging
import re
from typing import Dict, Any, List, Optional
from langchain_core.messages import BaseMessage, HumanMessage, AIMessage
logger = logging.getLogger(__name__)
def _extract_text(content) -> str:
"""Extract plain text từ msg.content."""
if isinstance(content, str):
return content
if isinstance(content, list):
parts = []
for item in content:
if isinstance(item, dict) and item.get("type") == "text":
parts.append(item.get("text", ""))
elif isinstance(item, str):
parts.append(item)
return "\n".join(parts) if parts else str(content)
return str(content or "")
_SKU_REGEX = re.compile(r"\b\d[A-Z0-9]{5,}(?:-[A-Z0-9]{2,}){0,2}\b", re.IGNORECASE)
def _dedupe_preserve_order(items: list[str]) -> list[str]:
seen: set[str] = set()
out: list[str] = []
for raw in items:
val = str(raw or "").upper().strip()
if not val or val in seen:
continue
seen.add(val)
out.append(val)
return out
def _extract_skus_from_text(text: str | None) -> list[str]:
if not text:
return []
matches = _SKU_REGEX.findall(text)
return _dedupe_preserve_order([m for m in matches if m])
def _format_history(messages: list[BaseMessage], max_turns: int = 10) -> str:
"""Format recent messages cho context."""
lines = []
human_count = 0
for msg in reversed(messages):
if isinstance(msg, HumanMessage):
lines.insert(0, f"Khách: {_extract_text(msg.content)[:500]}")
human_count += 1
elif isinstance(msg, AIMessage) and not (hasattr(msg, "tool_calls") and msg.tool_calls):
lines.insert(0, f"Bot: {_extract_text(msg.content)[:500]}")
if human_count >= max_turns:
break
return "\n".join(lines) if lines else "Chưa có lịch sử."
def _format_tool_result(tool_result: str | None) -> str:
"""Format kết quả từ tool thành text xịn cho Stylist."""
if not tool_result:
return ""
try:
parsed = json.loads(tool_result)
if parsed.get("status") == "success":
products = parsed.get("products", [])
if products:
dedup = {}
for p in products:
sku = str(p.get("sku_code") or p.get("sku") or p.get("internal_ref_code") or "").upper().strip()
if sku and sku not in dedup:
dedup[sku] = p
unique_products = list(dedup.values())
lines = [f"✅ Tìm thấy {len(unique_products)} sản phẩm:"]
for p in unique_products[:8]:
price = p.get("price", 0)
discount = p.get("discount", "")
line = f"- [{p.get('sku', '')}] {p.get('name', '')} | Giá: {price:,}đ"
if discount:
line += f" ({discount})"
if p.get("sizes"):
line += f" | Sizes: {p['sizes']}"
desc = (p.get("description") or "").strip()
if desc:
line += f"\n 📝 {desc[:200]}"
outfit = p.get("outfit_recommendations")
if outfit and isinstance(outfit, list):
outfit_lines = []
for o in outfit[:3]:
outfit_lines.append(
f"{o.get('role', 'match')}: {o.get('match_product_name')} ({o.get('match_product_code')})"
)
if outfit_lines:
line += f"\n 👗 outfit_recommendations: {', '.join(outfit_lines)}"
lines.append(line)
return "\n".join(lines)
return parsed.get("message", "Không tìm thấy sản phẩm phù hợp.")
content = parsed.get("content") or parsed.get("message") or str(parsed)
return str(content)[:800]
except Exception:
return (tool_result or "")[:500]
def parse_llm_json(text: str) -> Dict[str, Any]:
"""Senior Parser: Cleanly extract JSON from LLM response."""
text = text.strip()
if text.startswith('```json'): text = text[7:]
if text.startswith('```'): text = text[3:]
if text.endswith('```'): text = text[:-3]
text = text.strip()
start = text.find("{")
end = text.rfind("}")
if start != -1 and end != -1:
text = text[start : end + 1]
try:
return json.loads(text)
except json.JSONDecodeError:
logger.error(f"❌ Failed to parse JSON: {text[:200]}")
raise
backend/agent/prompt_module/push_production.py/push_prod.py deleted 100644 → 0
View file @ f9b10ebe
"""
Push ALL prompts (modules + tools) to PRODUCTION Langfuse.
Usage: py push_prod.py
"""
import os
import sys
sys.path.insert(0, os.path.join(os.path.dirname(__file__), "..", ".."))
sys.stdout.reconfigure(encoding="utf-8")
# ---- PRODUCTION Langfuse credentials ----
os.environ["LANGFUSE_SECRET_KEY"] = "sk-lf-bd0c7202-a7f4-4399-a36e-65ebe2e35104"
os.environ["LANGFUSE_PUBLIC_KEY"] = "pk-lf-2bb249b1-77e3-4309-8954-312b8fb2fff9"
os.environ["LANGFUSE_BASE_URL"] = "http://172.16.2.207:3009"
from langfuse import Langfuse
BASE_DIR = os.path.dirname(os.path.abspath(__file__))
PROMPT_DIR = os.path.dirname(BASE_DIR) # prompt_module/ (parent of push_production.py/)
TOOL_DIR = os.path.join(os.path.dirname(PROMPT_DIR), "tool_prompts")
# ---- Prompt modules ----
PROMPT_MODULES = [
("02_rules.txt", "canifa-02-rules", ["canifa", "system-core"]),
("03_context.txt", "canifa-03-context", ["canifa", "system-core"]),
("04a_sales_core.txt", "canifa-04a-sales-core", ["canifa", "system-sales"]),
("04b_sales_thaomai.txt", "canifa-04b-sales-thaomai", ["canifa", "system-sales"]),
("04c_sales_upsell.txt", "canifa-04c-sales-upsell", ["canifa", "system-sales"]),
("04d_sales_urgency.txt", "canifa-04d-sales-urgency", ["canifa", "system-sales"]),
("05_tool_routing.txt", "canifa-05-tool-routing", ["canifa", "system-core"]),
("05b_tool_results.txt", "canifa-05b-tool-results", ["canifa", "system-core"]),
("05c_comparison.txt", "canifa-05c-comparison", ["canifa", "system-core"]),
("06_user_insight.txt", "canifa-06-user-insight", ["canifa", "system-core"]),
("07_output_format.txt", "canifa-07-output-format", ["canifa", "system-core"]),
]
CORE_FILE = "01_core.txt"
CORE_PROMPT_NAME = "canifa-stylist-system-prompt"
SEASON_PROMPT_NAME = "canifa-08-season"
DEFAULT_SEASON_CONTENT = """## HƯỚNG DẪN TƯ VẤN THEO MÙA / EVENT
**Thời điểm hiện tại:** Tháng 3/2026 — Mùa Xuân, chuyển giao Đông → Hè
**Ưu tiên sản phẩm mùa này:**
- Áo khoác nhẹ, cardigan (trời se lạnh buổi sáng/tối)
- Áo phông, áo thun (ban ngày ấm)
- Sơ mi dài tay (đi làm)
- Quần jeans, quần kaki (đa năng)
**Khi khách hỏi chung chung ("có gì hot?", "gợi ý đi"):**
→ Ưu tiên giới thiệu sản phẩm phù hợp thời tiết hiện tại
→ Nhắc sale/khuyến mãi nếu có
**Event đang diễn ra:**
- (Marketing cập nhật event tại đây)
"""
# ---- Tool prompts ----
TOOL_PROMPTS = [
("data_retrieval_tool.txt", "canifa-tool-data-retrieval", ["canifa", "tool-prompt"]),
("brand_knowledge_tool.txt", "canifa-tool-brand-knowledge", ["canifa", "tool-prompt"]),
("check_is_stock.txt", "canifa-tool-check-is-stock", ["canifa", "tool-prompt"]),
("store_search_tool.txt", "canifa-tool-store-search", ["canifa", "tool-prompt"]),
("promotion_canifa_tool.txt", "canifa-tool-promotion", ["canifa", "tool-prompt"]),
]
def read_file(directory: str, filename: str) -> str:
with open(os.path.join(directory, filename), "r", encoding="utf-8") as f:
return f.read()
def main():
print("🚀 PUSH ALL TO PRODUCTION LANGFUSE")
print(f" URL: {os.environ['LANGFUSE_BASE_URL']}")
print()
lf = Langfuse()
# ── Step 1: Prompt modules ──
print("=" * 60)
print("STEP 1: Prompt Modules")
print("=" * 60)
for filename, name, tags in PROMPT_MODULES:
content = read_file(PROMPT_DIR, filename)
lf.create_prompt(name=name, prompt=content, labels=["production"], tags=tags, type="text")
print(f" ✅ {filename:30s} → {name} ({len(content):,} chars)")
# Season
lf.create_prompt(name=SEASON_PROMPT_NAME, prompt=DEFAULT_SEASON_CONTENT,
labels=["production"], tags=["canifa", "system-addon"], type="text")
print(f" ✅ {'(season)':30s} → {SEASON_PROMPT_NAME}")
# Core (with composable references)
core_content = read_file(PROMPT_DIR, CORE_FILE)
refs = "\n".join(f"@@@langfusePrompt:name={n}|label=production@@@" for _, n, _ in PROMPT_MODULES)
refs += f"\n@@@langfusePrompt:name={SEASON_PROMPT_NAME}|label=production@@@"
composed = core_content + "\n" + refs + "\n"
lf.create_prompt(name=CORE_PROMPT_NAME, prompt=composed,
labels=["production"], tags=["canifa", "system-prompt"], type="text")
print(f" ✅ {'01_core.txt (composed)':30s} → {CORE_PROMPT_NAME}")
# ── Step 2: Tool prompts ──
print("\n" + "=" * 60)
print("STEP 2: Tool Prompts")
print("=" * 60)
for filename, name, tags in TOOL_PROMPTS:
filepath = os.path.join(TOOL_DIR, filename)
if not os.path.exists(filepath):
print(f" ⚠️ SKIP {filename} — not found")
continue
content = read_file(TOOL_DIR, filename)
lf.create_prompt(name=name, prompt=content, labels=["production"], tags=tags, type="text")
print(f" ✅ {filename:35s} → {name} ({len(content):,} chars)")
# ── Step 3: Verify ──
print("\n" + "=" * 60)
print("STEP 3: Verification")
print("=" * 60)
prompt = lf.get_prompt(CORE_PROMPT_NAME, label="production", cache_ttl_seconds=0)
print(f" Assembled prompt: {len(prompt.prompt):,} chars")
checks = [
("01 Core", "C-stylist"),
("02 Rules", "QUY TẮC TRUNG THỰC"),
("03 Context", "CONTEXT AWARENESS"),
("04a Sales", "PHONG CÁCH TƯ VẤN"),
("05 Tool Route", "KHI NÀO GỌI TOOL"),
("05b Results", "XỬ LÝ KẾT QUẢ TOOL"),
("05c Compare", "SO SÁNH"),
("06 Insight", "USER INSIGHT 2.0"),
("07 Output", "FORMAT ĐẦU RA"),
("08 Season", "HƯỚNG DẪN TƯ VẤN THEO MÙA"),
]
all_ok = True
for label, keyword in checks:
found = keyword in prompt.prompt
print(f" {'✅' if found else '❌'} {label}: '{keyword}'")
if not found:
all_ok = False
for filename, name, _ in TOOL_PROMPTS:
if not os.path.exists(os.path.join(TOOL_DIR, filename)):
continue
try:
p = lf.get_prompt(name, label="production", cache_ttl_seconds=0)
print(f" ✅ {name} ({len(p.prompt):,} chars)")
except Exception as e:
print(f" ❌ {name} — {e}")
all_ok = False
lf.flush()
print(f"\n{'🎉 ALL DONE!' if all_ok else '⚠️ Done with warnings'}")
if __name__ == "__main__":
main()
backend/agent/prompt_module/push_production.py/push_producton.py deleted 100644 → 0
View file @ f9b10ebe
"""Push all prompts (system + tools) to Langfuse Production"""
import os
import sys
import subprocess
# --- 1. SET ENVIRONMENT VARIABLES FOR PRODUCTION ---
os.environ["LANGFUSE_SECRET_KEY"] = "sk-lf-b20df146-732b-47cb-9669-47983906eb93"
os.environ["LANGFUSE_PUBLIC_KEY"] = "pk-lf-611d32d6-e8c7-4e06-b5b5-c3486513a4a2"
os.environ["LANGFUSE_BASE_URL"] = "http://172.16.2.207:3009"
# We must import Langfuse AFTER setting os.environ, or initialize explicitly
from langfuse import Langfuse
def push_tools(lf):
print("\n" + "="*60)
print("🚀 PUSHING TOOL PROMPTS TO PRODUCTION")
print("="*60)
TOOL_PROMPTS_DIR = r"d:\cnf\chatbot_canifa\backend\agent\tool_prompts"
# Matching the exact names used in backend/agent/prompt_utils.py
TOOL_FILES = [
("brand_knowledge_tool.txt", "canifa-tool-brand-knowledge", ["canifa", "tool-prompt"]),
("check_is_stock.txt", "canifa-tool-check-stock", ["canifa", "tool-prompt"]),
("data_retrieval_tool.txt", "canifa-tool-data-retrieval", ["canifa", "tool-prompt"]),
("promotion_canifa_tool.txt","canifa-tool-promotion", ["canifa", "tool-prompt"]),
("store_search_tool.txt", "canifa-tool-store-search", ["canifa", "tool-prompt"]),
]
for filename, langfuse_name, tags in TOOL_FILES:
path = os.path.join(TOOL_PROMPTS_DIR, filename)
with open(path, "r", encoding="utf-8") as f:
content = f.read()
lf.create_prompt(
name=langfuse_name,
prompt=content,
labels=["production"],
tags=tags,
type="text",
)
print(f" ✅ {filename:30s} → {langfuse_name} ({len(content):,} chars)")
lf.flush()
print("✅ Tool prompts pushed successfully!\n")
def push_system_prompts():
print("="*60)
print("🚀 PUSHING SYSTEM PROMPTS TO PRODUCTION")
print("="*60)
# We can just call the existing push_modules_to_langfuse.py script
# since we already exported the env vars, it will pick up the prod credentials.
script_dir = r"d:\cnf\chatbot_canifa\backend\agent\prompt_module"
script_path = os.path.join(script_dir, "push_modules_to_langfuse.py")
env = os.environ.copy()
subprocess.run([sys.executable, script_path], cwd=script_dir, env=env, check=True)
if __name__ == "__main__":
sys.stdout.reconfigure(encoding="utf-8")
print("🌟 DEPLOYING TO C-STYLIST PRODUCTION 🌟")
print(f"Target: {os.environ.get('LANGFUSE_BASE_URL')}")
# 1. Push System Prompts
push_system_prompts()
# 2. Push Tool Prompts
lf = Langfuse()
push_tools(lf)
print("🎉 ALL DEPLOYMENTS TO PRODUCTION FINISHED SUCCESSFULLY!")
backend/agent/prompt_module/stylist_pro_prompts.py 0 → 100644
View file @ 94484304
"""
Prompts for StylistPro Agent (Architecture 2.0).
Combining legacy business logic with modern XML structural tagging for maximum LLM performance.
"""
# ==============================================================================
# 1. CLASSIFIER PROMPT (The Router)
# ==============================================================================
STYLIST_PRO_CLASSIFIER_PROMPT = """\
<system_role>
Bạn là AI Classifier (Router) cấp cao của CANIFA — thương hiệu thời trang gia đình hàng đầu Việt Nam.
Nhiệm vụ duy nhất: Phân tích intent khách hàng và quyết định lộ trình xử lý (Chọn Tool).
</system_role>
<tool_inventory>
1. `lead_search_tool`: Tìm kiếm sản phẩm (Sử dụng cho 90% trường hợp tìm đồ, combo, giá, màu).
2. `check_is_stock`: Kiểm tra size, tồn kho, giá cụ thể của 1 SKU.
3. `canifa_store_search`: Tìm địa chỉ cửa hàng.
4. `canifa_get_promotions`: Tra cứu các chương trình khuyến mãi hiện có.
5. `canifa_knowledge_search`: Tra cứu kiến thức thương hiệu, chính sách đổi trả, chất liệu.
</tool_inventory>
<search_philosophy_dual_lane>
Khi dùng `lead_search_tool`, bạn BẮT BUỘC phân tách:
- **Lane 1 (Literal)**: Trích xuất raw_text. Bỏ từ thừa (tìm, mua, cái...), sửa lỗi typo (nghọ nghiêng -> ngọ nghiêng).
- **Lane 2 (Inferred)**: Suy luận SQL Filters (product_line_vn, gender_target, age_group, tags, keywords).
</search_philosophy_dual_lane>
<vietnam_holidays_mapping>
- 30/4 & 2/9: master_color="Màu đỏ", tags=["Màu đỏ", "Đi chơi / dạo phố"], keywords=["cờ đỏ sao vàng"].
- Tết Nguyên Đán: master_color="Màu đỏ/Màu vàng", keywords=["Tết", "may mắn", "du xuân"], product_line_vn=["Váy liền", "Áo dài"].
- Valentine (14/2): master_color="Màu hồng", keywords=["Valentine", "lãng mạn"].
</vietnam_holidays_mapping>
<mapping_logic_rules>
1. Thời tiết: "Trời mưa" -> Áo khoác gió, trượt nước. "Nắng nóng" -> Áo chống nắng, UV.
2. Dịp: "Đi tiệc" -> Váy liền, Áo kiểu, Sang trọng. "Đi làm" -> Áo sơ mi, Polo, Quần Khaki (Adult).
3. Đồ lót: "Sịp/chip" -> Quần lót, Áo lót. (Tool tự tìm gender, bạn chỉ cần set product_line).
4. Combo: Khách hỏi "đồ", "set", "phối" -> BẮT BUỘC gọi lead_search_tool. RESET product_line cũ từ insight.
</mapping_logic_rules>
<inheritance_rules>
- Kế thừa Giới tính/Đối tượng từ <user_insight> (VD: Target là Nam -> gender_target="men").
- RESET hoàn toàn product_line_vn theo câu hỏi mới nhất.
</inheritance_rules>
<guardrails>
Nghiêm túc chặn các chủ đề không liên quan (Off-topic):
1. Chính trị, tôn giáo, phân biệt chủng tộc, bạo lực, nội dung 18+.
2. So sánh trực diện hoặc tâng bốc đối thủ (Uniqlo, Ivy Moda, Yody, v.v.).
3. Các câu hỏi linh tinh không liên quan đến thời trang, phong cách sống, hoặc Canifa (Ví dụ: "Giá vàng", "Cách nấu phở", "Review phim", "Code Python").
4. Khi gặp off-topic: Set `tool_name: null`, `ai_response`: "Dạ, hiện tại em là trợ lý thời trang của Canifa nên em chỉ có thể hỗ trợ mình về phối đồ và sản phẩm thôi ạ. Anh/Chị cần em tư vấn bộ đồ nào cho hôm nay không ạ?"
</guardrails>
<output_format>
Trả về DUY NHẤT JSON:
{
"reasoning": "giải thích ngắn gọn tại sao chọn tool này hoặc tại sao chặn",
"tool_name": "tên_tool" hoặc null,
"search_args": {
"literal": {"raw_text": "..."},
"inferred": {"product_line_vn": [...], "gender_target": "...", "tags": [...], "keywords": [...], ...}
},
"ai_response": "chào hỏi xã giao HOẶC từ chối off-topic (chỉ dùng khi tool_name=null)",
"product_ids": []
}
</output_format>
"""
# ==============================================================================
# 2. STYLIST PROMPT (The Expert)
# ==============================================================================
STYLIST_PRO_RESPONDER_PROMPT = """\
<system_role>
Bạn là Chuyên gia Thời trang (Stylist Pro) của CANIFA. Bạn tư vấn dựa trên dữ liệu sản phẩm thực và triết lý thời trang của hãng.
</system_role>
<styling_philosophy>
1. GỢI Ý NGAY, KHÔNG HỎI LẠI: Luôn đưa ra lựa chọn. Nếu chưa biết giới tính, gợi ý cả 2 phương án.
2. COMBO / OUTFIT (CRITICAL):
- Nếu khách hỏi "đồ", "set", "phối", "mặc gì" -> PHẢI tạo outfit hoàn chỉnh (Áo + Quần/Váy).
- Sử dụng `outfit_recommendations` từ tool result để bốc đúng món phối.
- BẮT BUỘC nhét SKU món phối vào `product_ids`.
3. KIỂM TRA SỰ PHÙ HỢP: Đừng gợi ý đồ đông cho mùa hè, đừng gợi ý đồ đi làm là áo hoạt hình.
4. KHÔNG TRẦN TÌNH THUẬT TOÁN: Đừng nói "Hệ thống trả về...", hãy nói "Canifa đang có sẵn...".
</styling_philosophy>
<user_memory_update>
Cập nhật 12 trường Insight:
- STAGE: BROWSE (tìm kiếm) -> COMPARE (hỏi sâu tính năng/giá) -> DECIDE (chốt đơn/hỏi size).
- LATEST_PRODUCT_INTEREST: Ghi rõ Tên + SKU sản phẩm vừa tư vấn chính.
- SUMMARY_HISTORY: Tóm tắt các bước đã tư vấn.
</user_memory_update>
<formatting_rules>
- SKU sản phẩm phải bọc trong ngoặc tròn, ví dụ: (6TS25S001).
- MỌI SKU nhắc đến trong text BẮT BUỘC phải nằm trong mảng product_ids.
- Tối đa 250 từ. Không dán URL.
</formatting_rules>
<output_format>
Trả về DUY NHẤT JSON:
{{
"ai_response": "nội dung tư vấn chuyên nghiệp",
"product_ids": ["SKU1", "SKU2", ...],
"user_insight": {{ ...updated 12 fields... }}
}}
</output_format>
<context_injection>
{stage_injection}
</context_injection>
"""
def format_stylist_pro_injection(insight: dict) -> str:
"""Format InsightJSON dict thành đoạn XML context cho prompt."""
template = """
<user_memory>
- User Type: {user}
- Target: {target}
- Goal: {goal}
- Constraints: {constraints}
- Stage: {stage} (Level {stage_num})
- Tone: {tone}
- Latest Interest: {latest}
- History Summary: {summary}
</user_memory>
"""
return template.format(
user=insight.get("USER", "Chưa rõ"),
target=insight.get("TARGET", "Chưa rõ"),
goal=insight.get("GOAL", "Chưa rõ"),
constraints=insight.get("CONSTRAINTS") or "Không",
stage=insight.get("STAGE", "BROWSE"),
stage_num=insight.get("STAGE_NUM", 1),
tone=insight.get("TONE", "Friendly"),
latest=insight.get("LATEST_PRODUCT_INTEREST") or "Chưa có",
summary=insight.get("SUMMARY_HISTORY") or "Chưa có"
)
backend/agent/stylist_pro_graph.py 0 → 100644
View file @ 94484304
import logging
import json
import time
from typing import Dict, Any, List
from langgraph.graph import StateGraph, END
from langchain_core.messages import HumanMessage, AIMessage, BaseMessage
from langchain_core.runnables import RunnableConfig
# New Modular Imports
from backend.agent.nodes.models_node import StylistProState
from backend.agent.nodes.classifier_node import classifier_node
from backend.agent.nodes.stylist_node import stylist_node
from backend.agent.nodes.utils import _extract_text
# StarRocks connection for Enrichment
from agent.lead_stage_agent.product_search_engine import SELECT_COLUMNS, TABLE_NAME
from common.starrocks_connection import get_db_connection
logger = logging.getLogger(__name__)
def route_after_classifier(state: StylistProState) -> str:
"""Điều hướng dựa trên early_exit."""
if state.get("early_exit"):
return END
return "stylist"
class StylistProGraph:
"""
Architect Stylist Graph (Version 2.0).
Modular Architecture: Classifier -> Stylist -> END
"""
def __init__(self):
self._compiled = None
def build(self):
if self._compiled:
return self._compiled
workflow = StateGraph(StylistProState)
workflow.add_node("classifier", classifier_node)
workflow.add_node("stylist", stylist_node)
workflow.set_entry_point("classifier")
workflow.add_conditional_edges(
"classifier",
route_after_classifier,
{
"stylist": "stylist",
END: END
}
)
workflow.add_edge("stylist", END)
self._compiled = workflow.compile()
logger.info("✅ StylistProGraph compiled (Modular)")
return self._compiled
async def chat(
self,
user_message: str,
user_insight: str | None = None,
history: List[BaseMessage] | None = None,
config: RunnableConfig | None = None,
) -> Dict[str, Any]:
"""
Main entry point with StarRocks Product Enrichment.
"""
start_time = time.time()
graph = self.build()
messages = history.copy() if history else []
messages.append(HumanMessage(content=user_message))
initial_state = {
"messages": messages,
"user_insight": user_insight,
"tool_result": None,
"tool_name_used": None,
"updated_insight": None,
"lead_stage": None,
"stage_injection": None,
"product_ids": [],
"early_exit": False,
"diagnostics": [],
}
result = await graph.ainvoke(initial_state, config=config)
elapsed_ms = (time.time() - start_time) * 1000
# Extract AI Response
ai_response = ""
for msg in reversed(result["messages"]):
if isinstance(msg, AIMessage):
ai_response = _extract_text(msg.content)
break
# --- PRODUCT ENRICHMENT ---
final_products = []
ai_product_ids = result.get("product_ids", [])
tool_result_raw = result.get("tool_result")
product_dict = {}
if tool_result_raw:
try:
parsed = json.loads(tool_result_raw)
if parsed.get("status") == "success":
all_products = parsed.get("products", [])
for p in all_products:
sku = str(p.get("magento_ref_code") or p.get("sku") or "").upper().strip()
if sku:
if "sku" not in p: p["sku"] = sku
product_dict[sku] = p
recs = p.get("outfit_recommendations", [])
if isinstance(recs, list):
for r in recs:
r_sku = str(r.get("match_product_code") or r.get("sku") or "").upper().strip()
if r_sku:
if "sku" not in r: r["sku"] = r_sku
product_dict[r_sku] = r
except Exception: pass
# Enrich missing IDs via StarRocks
wanted_ids = [pid.upper().strip() for pid in ai_product_ids if pid]
missing_ids = [pid for pid in wanted_ids if pid not in product_dict or not product_dict[pid].get("image")]
if missing_ids:
try:
db = get_db_connection()
skus = list(dict.fromkeys(missing_ids))
base_skus = list(dict.fromkeys([s.split("-")[0] for s in skus]))
keys = list(dict.fromkeys(skus + base_skus))
placeholders = ",".join(["%s"] * len(keys))
sql = f"SELECT {SELECT_COLUMNS} FROM {TABLE_NAME} WHERE UPPER(magento_ref_code) IN ({placeholders}) OR UPPER(internal_ref_code) IN ({placeholders}) LIMIT 200"
rows = await db.execute_query_async(sql, params=tuple(keys + keys))
for r in rows or []:
card = {
"sku": (r.get("magento_ref_code") or r.get("internal_ref_code") or "").strip(),
"name": r.get("product_name", ""),
"price": int(r.get("sale_price") or 0),
"original_price": int(r.get("original_price") or 0),
"image": r.get("product_image_url_thumbnail", ""),
"url": r.get("product_web_url", ""),
"sizes": r.get("size_scale", ""),
"gender": r.get("gender_by_product", ""),
"product_line": r.get("product_line_vn", ""),
}
product_dict[card["sku"].upper()] = card
except Exception as e:
logger.warning(f"⚠️ Enrichment failed: {e}")
seen = set()
for pid in wanted_ids:
if pid in product_dict and pid not in seen:
final_products.append(product_dict[pid])
seen.add(pid)
return {
"response": ai_response,
"elapsed_ms": round(elapsed_ms, 2),
"lead_stage": result.get("lead_stage"),
"updated_insight": result.get("updated_insight"),
"product_ids": ai_product_ids,
"products": final_products,
"pipeline": result.get("diagnostics", [])
}
# Singleton
stylist_pro_graph = StylistProGraph()
backend/agent/tools/__init__.py
View file @ 94484304
"""
Tools Package
Export tool và factory function
Tools package exports.
"""
from .data_retrieval_tool import data_retrieval_tool
from .get_tools import get_all_tools
from .promotion_canifa_tool import canifa_get_promotions
def get_all_tools():
from .get_tools import get_all_tools as _get_all_tools
return _get_all_tools()
def __getattr__(name: str):
if name == "canifa_get_promotions":
from .promotion_canifa_tool import canifa_get_promotions
return canifa_get_promotions
raise AttributeError(name)
__all__ = ["data_retrieval_tool", "get_all_tools", "canifa_get_promotions"]
backend/agent/tools/brand_knowledge_tool.py
View file @ 94484304
......@@ -21,20 +21,19 @@ class KnowledgeSearchInput(BaseModel):
@tool("canifa_knowledge_search", args_schema=KnowledgeSearchInput)
async def canifa_knowledge_search(query: str) -> str:
"""
(Placeholder docstring - Actual prompt is loaded from file)
Tra cứu thông tin thương hiệu Canifa (chính sách, KHTT, bảng size...).
KHÔNG dùng để tìm cửa hàng - dùng canifa_store_search.
"""
logger.info(f"[Semantic Search] Brand Knowledge query: {query}")
try:
# 1. Tạo embedding cho câu hỏi (cached 24h trong Redis)
# 1. Tạo embedding cho câu hỏi
query_vector = await create_embedding_async(query)
if not query_vector:
return "Xin lỗi, tôi gặp sự cố khi xử lý thông tin. Vui lòng thử lại sau."
v_str = json.dumps(query_vector)
# 2. SINGLE QUERY: Tìm top K chunks → lấy full document theo title (1 round-trip)
# 2. Tìm kiếm vector trên StarRocks
sql = f"""
WITH top_chunks AS (
SELECT CAST(json_query(metadata, '$.title') AS VARCHAR) AS title
......@@ -53,46 +52,25 @@ async def canifa_knowledge_search(query: str) -> str:
results = await sr.execute_query_async(sql)
if not results:
logger.warning(f"No knowledge data found in DB for query: {query}")
return "Hiện tại tôi chưa tìm thấy thông tin chính xác về nội dung này trong hệ thống kiến thức của Canifa. Bạn có thể liên hệ hotline 1800 6061 để được hỗ trợ trực tiếp."
return "Hiện tại tôi chưa tìm thấy thông tin chính xác về nội dung này trong hệ thống kiến thức của Canifa."
# 3. Phân nhóm nội dung theo title
docs_by_title: dict[str, list[str]] = {}
for res in results:
try:
meta_obj = (
json.loads(res.get("metadata", "{}"))
if isinstance(res.get("metadata"), str)
else res.get("metadata", {})
)
meta_obj = json.loads(res.get("metadata", "{}")) if isinstance(res.get("metadata"), str) else res.get("metadata", {})
title = meta_obj.get("title") if isinstance(meta_obj, dict) else None
content = res.get("content", "").strip()
if title and content:
docs_by_title.setdefault(title, []).append(content)
except Exception as e:
logger.debug(f"Skip invalid row while grouping by title: {e}")
if not docs_by_title:
# Fallback: trả raw content nếu không parse được title
knowledge_texts = [res.get("content", "") for res in results]
return "\n\n---\n\n".join(knowledge_texts)
# 4. Tổng hợp kết quả
final_blocks = []
for title, contents in docs_by_title.items():
if contents:
full_text = "\n\n".join(contents)
final_blocks.append(f"=== TÀI LIỆU: {title} ===\n{full_text}")
final_response = "\n\n".join(final_blocks)
logger.info(f"Parent Document Retrieval complete. Loaded full context for {len(final_blocks)} topics.")
return final_response
return "\n\n".join(final_blocks)
except Exception as e:
logger.error(f"Error in canifa_knowledge_search: {e}")
return "Tôi đang gặp khó khăn khi truy cập kho kiến thức. Bạn muốn hỏi về sản phẩm gì khác không?"
return "Tôi đang gặp khó khăn khi truy cập kho kiến thức."
canifa_knowledge_search.__doc__ = read_tool_prompt("brand_knowledge_tool") or canifa_knowledge_search.__doc__
backend/agent/tools/data_retrieval_tool.py
View file @ 94484304
"""
CANIFA Data Retrieval Tool - Tối giản cho Agentic Workflow.
Hỗ trợ Hybrid Search: Semantic (Vector) + Metadata Filter.
Tracing:
- LangChain CallbackHandler auto-traces tool call (name, input, output)
- Manual span 'db-query-search' in _execute_single_search for SQL timing & product count
(nested under parent LangChain trace via start_as_current_observation)
"""
import asyncio
import json
import logging
import time
import unicodedata
import re
import sys
from pathlib import Path
from typing import Any, Dict, List, Optional, Union
from langchain_core.runnables import RunnableConfig
from langchain_core.tools import tool
from langfuse import get_client as get_langfuse
from pydantic import BaseModel, Field
from agent.helper import format_product_results
from agent.prompt_utils import read_tool_prompt
from agent.tools.product_search_helpers import build_starrocks_query
from common.starrocks_connection import get_db_connection
from .tool_module import SearchEngine as LegacySearchEngine
# Setup Logger
logger = logging.getLogger(__name__)
PROJECT_ROOT = Path(__file__).resolve().parents[3]
if str(PROJECT_ROOT) not in sys.path:
sys.path.append(str(PROJECT_ROOT))
class SearchItem(BaseModel):
model_config = {"extra": "forbid"} # STRICT MODE
# ====== SEMANTIC SEARCH ======
description: str = Field(
description=(
"Mô tả sản phẩm cho semantic search. "
"FORMAT: 'product_name: [tên]. description_text: [mô tả]. "
"material_group: [chất liệu]. style: [phong cách]. fitting: [dáng]. "
"form_sleeve: [tay]. form_neckline: [cổ]. product_line_vn: [dòng SP]'. "
"🚨 product_name = CHÍNH XÁC lời user nói, TUYỆT ĐỐI KHÔNG tự đổi sang tên khác. "
"VD: user nói 'áo ngọ nguậy' → product_name: Áo ngọ nguậy (KHÔNG đổi thành 'Áo phông'!). "
"VD: user nói 'áo cá sấu' → product_name: Áo cá sấu (KHÔNG đổi!). "
"Chỉ chuẩn hóa từ đồng nghĩa RÕ RÀNG: 'áo thun'→'Áo phông', 'quần bò'→'Quần jean'. "
"⚠️ KHÔNG đưa master_color vào description — dùng field master_color riêng! "
"product_line_vn trong description = broad category (Áo phông, Áo Polo, Quần jean...). "
"VD: 'áo ngọ nguậy' → 'product_name: Áo ngọ nguậy. product_line_vn: Áo phông'. "
"VD: 'áo cá sấu polo' → 'product_name: Áo cá sấu polo. product_line_vn: Áo Polo'."
)
)
product_name: str | None = Field(
description="Tên sản phẩm. Chỉ điền khi khách cung cấp tên cụ thể.",
try:
from preference.lead_stage_agent.product_search_engine import (
InferredSearch as ProductInferredSearch,
LeadSearchInput,
LiteralSearch as ProductLiteralSearch,
ProductSearchEngine,
)
except Exception as exc: # pragma: no cover - exercised only when optional import is unavailable
ProductInferredSearch = None
ProductLiteralSearch = None
ProductSearchEngine = None
LeadSearchInput = None
logger.warning("ProductSearchEngine import failed; legacy SearchEngine fallback will be used: %s", exc)
# ====== SKU LOOKUP ======
magento_ref_code: str | None = Field(
description="Mã sản phẩm chính xác (SKU). Chỉ điền khi khách cung cấp mã cụ thể.",
)
# ====== SQL HARD FILTERS (lọc trực tiếp trong DB query) ======
gender_by_product: str | None = Field(
description="[SQL FILTER] Giới tính. GIÁ TRỊ HỢP LỆ: women, men, boy, girl, unisex, newborn",
)
age_by_product: str | None = Field(
description="[SQL FILTER] Độ tuổi. GIÁ TRỊ HỢP LỆ: adult, kid, others",
)
master_color: str | None = Field(
description=(
"[SQL FILTER] Màu sắc sản phẩm. Gửi CHÍNH XÁC từ khách nói (VD: 'trắng', 'đen', 'xanh', 'đỏ'). "
"Tool sẽ tự match trong DB bằng LIKE. KHÔNG đưa màu vào description!"
),
)
price_min: int | None = Field(
description="[SQL FILTER] Giá thấp nhất (VND)",
)
price_max: int | None = Field(
description="[SQL FILTER] Giá cao nhất (VND)",
)
discount_min: int | None = Field(
description="[SQL FILTER] % giảm giá tối thiểu. VD: 50 → chỉ lấy SP giảm >= 50%. Nếu khách hỏi 'đồ sale' chung → đặt = 1.",
)
discount_max: int | None = Field(
description="[SQL FILTER] % giảm giá tối đa. VD: 70 → chỉ lấy SP giảm <= 70%.",
)
discovery_mode: str | None = Field(
description=(
"[SQL FILTER] Chế độ khám phá. 'new' = hàng mới (is_new_product=1), "
"'best_seller' = bán chạy nhất (ORDER BY quantity_sold). "
"Chỉ dùng khi khách NÓI RÕ 'mới nhất'/'hàng mới'/'bán chạy'/'best seller'/'hot'. "
"Nếu khách KHÔNG nói → để null."
),
)
class LiteralSearch(BaseModel):
model_config = {"extra": "ignore"}
raw_text: str = Field(default="", description="Cau search nguyen van tu user.")
class MultiSearchParams(BaseModel):
model_config = {"extra": "forbid"} # STRICT MODE
searches: list[SearchItem] = Field(description="Danh sách các truy vấn tìm kiếm")
class InferredSearch(BaseModel):
model_config = {"extra": "ignore"}
def _normalize_text(value: str | None) -> str:
"""Normalize free text for robust case/accent-insensitive matching."""
if not value:
return ""
normalized = unicodedata.normalize("NFD", str(value))
no_diacritics = "".join(ch for ch in normalized if unicodedata.category(ch) != "Mn")
return " ".join(no_diacritics.lower().strip().split())
product_line_vn: List[str] = Field(default_factory=list, description="Dong san pham.")
gender_by_product: Optional[str] = Field(default=None, description="Gioi tinh.")
age_by_product: Optional[str] = Field(default=None, description="Do tuoi.")
gender_target: Optional[str] = Field(default=None, description="Alias cu cua gender_by_product.")
age_group: Optional[str] = Field(default=None, description="Alias cu cua age_by_product.")
master_color: Optional[str] = Field(default=None, description="Mau sac.")
tags: List[str] = Field(default_factory=list, description="Intent tags: occ/style/fit/weather/function.")
keywords: List[str] = Field(default_factory=list, description="Tu khoa bo tro cho search.")
price_min: Optional[int] = Field(default=None, description="Gia thap nhat.")
price_max: Optional[int] = Field(default=None, description="Gia cao nhat.")
discount_min: Optional[int] = Field(default=None, description="% giam gia toi thieu.")
discount_max: Optional[int] = Field(default=None, description="% giam gia toi da.")
discovery_mode: Optional[str] = Field(default=None, description="new/best_seller.")
size: Optional[str] = Field(default=None, description="Size can tim.")
def _attach_variant_skus(formatted_products: list[dict], raw_products: list[dict]) -> None:
"""Attach full variant SKUs to grouped products for downstream stock lookup."""
by_base_sku: dict[str, list[dict]] = {}
class SearchItem(BaseModel):
"""
Supports both the current dual-lane schema and older flat tool calls.
Older endpoints still pass fields like query/product_name/product_line_vn directly.
"""
model_config = {"extra": "ignore"}
literal: LiteralSearch = Field(default_factory=LiteralSearch)
inferred: InferredSearch = Field(default_factory=InferredSearch)
magento_ref_code: Optional[str] = None
reasoning: Optional[str] = None
user_insight: Optional[Dict[str, Any]] = None
# Legacy flat fields.
query: Optional[str] = None
raw_text: Optional[str] = None
description: Optional[str] = None
product_name: Optional[str] = None
product_line_vn: Optional[Union[str, List[str]]] = None
gender_by_product: Optional[str] = None
age_by_product: Optional[str] = None
gender_target: Optional[str] = None
age_group: Optional[str] = None
master_color: Optional[str] = None
tags: Optional[Union[str, List[str]]] = None
keywords: Optional[Union[str, List[str]]] = None
price_min: Optional[int] = None
price_max: Optional[int] = None
discount_min: Optional[int] = None
discount_max: Optional[int] = None
discovery_mode: Optional[str] = None
size: Optional[str] = None
for row in raw_products:
base_sku = str(row.get("magento_ref_code"))
variant_sku = str(row.get("product_color_code") or "").strip()
if not base_sku or not variant_sku:
continue
by_base_sku.setdefault(base_sku, []).append(
{
"sku_color": variant_sku,
"color": row.get("master_color") or "",
"url": row.get("product_web_url") or "",
"thumbnail_image_url": row.get("product_image_url_thumbnail") or "",
"thumbnail": row.get("product_image_url_thumbnail") or "",
"price": int(row.get("original_price") or 0),
"sale_price": int(row.get("sale_price") or row.get("original_price") or 0),
}
)
class MultiSearchParams(BaseModel):
searches: List[SearchItem] = Field(..., description="Danh sach cac yeu cau tim kiem.")
reasoning: Optional[str] = None
user_insight: Optional[Dict[str, Any]] = None
for product in formatted_products:
base_sku = str(product.get("sku") or "").strip()
if not base_sku:
continue
variants = by_base_sku.get(base_sku, [])
if not variants:
continue
_PRODUCT_ENGINE = ProductSearchEngine() if ProductSearchEngine else None
deduped: list[dict] = []
seen_variant_skus: set[str] = set()
for variant in variants:
sku_color = str(variant.get("sku_color") or "").strip()
if not sku_color or sku_color in seen_variant_skus:
continue
seen_variant_skus.add(sku_color)
deduped.append(variant)
if deduped:
product["variants"] = deduped
def _coerce_list(value: Any) -> List[str]:
if value is None:
return []
if isinstance(value, list):
return [str(v).strip() for v in value if str(v).strip()]
if isinstance(value, tuple):
return [str(v).strip() for v in value if str(v).strip()]
text = str(value).strip()
if not text:
return []
if "/" in text:
return [part.strip() for part in text.split("/") if part.strip()]
return [text]
def _resolve_stock_skus(searches: list[SearchItem], products: list[dict]) -> list[str]:
"""Resolve full stock SKUs from returned flat products, expanding by size when possible."""
resolved: list[str] = []
seen: set[str] = set()
def _clean_legacy_description(description: Optional[str]) -> str:
if not description:
return ""
text = str(description)
text = re.sub(
r"\b(product_name|description_text|product_line_vn|material_group|style|fitting|form_sleeve|form_neckline|season)\s*:\s*",
" ",
text,
flags=re.IGNORECASE,
)
text = re.sub(r"[.;]+", " ", text)
return re.sub(r"\s+", " ", text).strip()
for item in searches:
base_code = str(item.magento_ref_code or "").strip().upper()
if not base_code:
continue
target_color = _normalize_text(item.master_color)
candidates = []
for product in products:
sku = str(product.get("sku") or "").strip().upper()
sku_color = str(product.get("sku_color") or "").strip().upper()
if not sku and not sku_color:
continue
if base_code and (sku == base_code or sku_color == base_code or sku.startswith(f"{base_code}-")):
candidates.append(product)
for product in candidates:
sku_color = str(product.get("sku_color") or "").strip()
product_color = _normalize_text(product.get("color"))
sizes = [str(size).strip().upper() for size in (product.get("sizes") or []) if str(size).strip()]
expanded_skus = [f"{sku_color}-{size}" for size in sizes] if sku_color and sizes else ([sku_color] if sku_color else [])
if target_color:
if target_color in product_color or product_color in target_color:
for full_sku in expanded_skus:
if full_sku and full_sku not in seen:
seen.add(full_sku)
resolved.append(full_sku)
def _first_text(*values: Any) -> str:
for value in values:
if value is None:
continue
text = str(value).strip()
if text:
return text
return ""
for full_sku in expanded_skus:
if full_sku and full_sku not in seen:
seen.add(full_sku)
resolved.append(full_sku)
return resolved
def _build_literal_text(item: SearchItem) -> str:
return _first_text(
item.literal.raw_text,
item.raw_text,
item.query,
item.product_name,
_clean_legacy_description(item.description),
)
async def _execute_single_search(
db, item: SearchItem, query_vector: list[float] | None = None
) -> tuple[list[dict], dict]:
"""
Thực thi một search query đơn lẻ (Async).
Creates a manual 'db-query-search' span under the current Langfuse context.
def _set_if_present(target: Dict[str, Any], key: str, value: Any) -> None:
if value is None:
return
if isinstance(value, list) and not value:
return
if value == "":
return
target[key] = value
def _build_inferred_payload(item: SearchItem) -> Dict[str, Any]:
data = item.inferred.model_dump(exclude_none=True)
if data.get("gender_target") and not data.get("gender_by_product"):
data["gender_by_product"] = data["gender_target"]
if data.get("age_group") and not data.get("age_by_product"):
data["age_by_product"] = data["age_group"]
flat_fields = [
"gender_by_product",
"age_by_product",
"gender_target",
"age_group",
"master_color",
"price_min",
"price_max",
"discount_min",
"discount_max",
"discovery_mode",
"size",
]
for field_name in flat_fields:
current = data.get(field_name)
if current not in (None, "", []):
continue
value = getattr(item, field_name, None)
_set_if_present(data, field_name, value)
if data.get("gender_target") and not data.get("gender_by_product"):
data["gender_by_product"] = data["gender_target"]
if data.get("age_group") and not data.get("age_by_product"):
data["age_by_product"] = data["age_group"]
product_lines = data.get("product_line_vn") or item.product_line_vn
data["product_line_vn"] = _coerce_list(product_lines)
data["tags"] = _coerce_list(data.get("tags") or item.tags)
data["keywords"] = _coerce_list(data.get("keywords") or item.keywords)
return {
"product_line_vn": data.get("product_line_vn", []),
"gender_by_product": data.get("gender_by_product"),
"age_by_product": data.get("age_by_product"),
"master_color": data.get("master_color"),
"tags": data.get("tags", []),
"keywords": data.get("keywords", []),
"price_min": data.get("price_min"),
"price_max": data.get("price_max"),
"discount_min": data.get("discount_min"),
"discount_max": data.get("discount_max"),
"discovery_mode": data.get("discovery_mode"),
"size": data.get("size"),
}
Returns:
Tuple of (products, filter_info)
"""
try:
short_query = (
(item.description[:60] + "...") if item.description and len(item.description) > 60 else item.description
)
logger.debug(
"_execute_single_search started, query=%r, code=%r",
short_query,
item.magento_ref_code,
)
# Timer: build query (sử dụng vector đã có hoặc build mới)
query_build_start = time.time()
sql, params = await build_starrocks_query(item, query_vector=query_vector)
query_build_time = (time.time() - query_build_start) * 1000 # Convert to ms
logger.debug("SQL built, length=%s, build_time_ms=%.2f", len(sql), query_build_time)
if not sql:
return [], {"fallback_used": False}
# Timer: execute DB query
db_start = time.time()
products = await db.execute_query_async(sql, params=params)
db_time = (time.time() - db_start) * 1000 # Convert to ms
logger.info(
"🔍 _execute_single_search done: query=%r, products=%s, build_ms=%.2f, db_ms=%.2f, total_ms=%.2f",
short_query,
len(products),
query_build_time,
db_time,
query_build_time + db_time,
def _build_lead_input(item: SearchItem) -> Any:
if not (LeadSearchInput and ProductLiteralSearch and ProductInferredSearch):
return None
return LeadSearchInput(
literal=ProductLiteralSearch(raw_text=_build_literal_text(item)),
inferred=ProductInferredSearch(**_build_inferred_payload(item)),
magento_ref_code=item.magento_ref_code,
reasoning=item.reasoning,
)
# Debug: Log first product to see fields
if products:
first_p = products[0]
logger.debug("🔍 [DEBUG] First product keys: %s", list(first_p.keys()))
logger.debug(
"🔍 [DEBUG] First product price: %s, sale_price: %s",
first_p.get("original_price"),
first_p.get("sale_price"),
def _augment_product(product: Dict[str, Any]) -> Dict[str, Any]:
p = dict(product)
sku = (
p.get("sku")
or p.get("magento_ref_code")
or p.get("internal_ref_code")
or p.get("code")
or ""
)
logger.debug(
"🔍 [DEBUG] size_scale: %r, description_text: %r",
first_p.get("size_scale"),
(first_p.get("description_text") or "")[:100],
name = p.get("name") or p.get("product_name") or p.get("magento_product_name") or ""
price = p.get("price", p.get("sale_price", p.get("sale_price_vnd", 0)))
original_price = p.get("original_price", p.get("price_vnd", price))
image = p.get("image") or p.get("product_image_url_thumbnail") or p.get("thumbnail_image_url") or ""
url = p.get("url") or p.get("product_web_url") or p.get("magento_url_key") or ""
p.setdefault("sku", sku)
p.setdefault("code", sku)
p.setdefault("magento_ref_code", sku)
p.setdefault("internal_ref_code", sku)
p.setdefault("name", name)
p.setdefault("product_name", name)
p.setdefault("magento_product_name", name)
p.setdefault("price", price)
p.setdefault("sale_price", price)
p.setdefault("sale_price_vnd", price)
p.setdefault("price_vnd", original_price)
p.setdefault("image", image)
p.setdefault("product_image_url_thumbnail", image)
p.setdefault("thumbnail_image_url", image)
p.setdefault("url", url)
p.setdefault("product_web_url", url)
p.setdefault("magento_url_key", url)
return p
def _dedup_products(products: List[Dict[str, Any]]) -> List[Dict[str, Any]]:
seen: set[str] = set()
deduped: List[Dict[str, Any]] = []
for product in products:
sku = str(product.get("sku") or product.get("magento_ref_code") or product.get("internal_ref_code") or "")
if sku and sku in seen:
continue
if sku:
seen.add(sku)
deduped.append(product)
return deduped
async def _run_product_search_engine(
item: SearchItem,
*,
reasoning: Optional[str],
user_insight: Optional[Dict[str, Any]],
) -> Dict[str, Any]:
if not _PRODUCT_ENGINE:
raise RuntimeError("ProductSearchEngine is not available")
req = _build_lead_input(item)
result = await _PRODUCT_ENGINE.search(
req,
reasoning=item.reasoning or reasoning,
user_insight=item.user_insight or user_insight,
)
products = [_augment_product(p) for p in result.get("products", [])]
result["products"] = products
result["engine_used"] = "product_search_engine"
return result
# ✅ Nest 'db-query-search' span dưới chat-request trace hiện tại trên Langfuse
try:
lf = get_langfuse()
if lf:
with lf.start_as_current_observation(
as_type="span",
name="db-query-search",
) as span:
span.update(
input={
"description": item.description[:200] if item.description else None,
"product_name": item.product_name,
"gender": item.gender_by_product,
"age": item.age_by_product,
"color": item.master_color,
"price_range": f"{item.price_min or ''}-{item.price_max or ''}",
"magento_ref_code": item.magento_ref_code,
"discovery_mode": item.discovery_mode,
},
output={
"product_count": len(products),
"build_ms": round(query_build_time, 2),
"db_ms": round(db_time, 2),
"total_ms": round(query_build_time + db_time, 2),
"sample_products": [
{
"name": p.get("product_name", "?"),
"code": p.get("magento_ref_code", "?"),
"price": float(p["sale_price"]) if p.get("sale_price") else None,
}
for p in products[:3]
]
if products
else [],
},
metadata={"sql_length": len(sql)},
)
except Exception as trace_err:
logger.warning("⚠️ Langfuse db-query-search span failed: %s", trace_err, exc_info=True)
formatted_products = format_product_results(products)
return formatted_products, {"fallback_used": False}
except Exception as e:
logger.exception("Single search error for item %r: %s", item, e)
return [], {"fallback_used": False, "error": str(e)}
async def _run_legacy_engine(item: SearchItem, *, use_sqlite: bool) -> Dict[str, Any]:
engine = LegacySearchEngine(use_sqlite=use_sqlite)
res = await engine.search(
literal=_build_literal_text(item),
inferred=_build_inferred_payload(item),
)
products = [_augment_product(p) for p in res.get("results", [])]
return {
"status": "success" if products else "no_results",
"count": len(products),
"tier": res.get("tier", 0),
"search_mode": res.get("message", "legacy_search"),
"message": res.get("message", ""),
"products": products,
"engine_used": "legacy_search_engine",
}
@tool(args_schema=MultiSearchParams)
async def data_retrieval_tool(searches: list[SearchItem]) -> str:
async def data_retrieval_tool(
searches: List[SearchItem],
reasoning: Optional[str] = None,
user_insight: Optional[Dict[str, Any]] = None,
config: RunnableConfig = None,
) -> str:
"""
Công cụ tìm kiếm sản phẩm CANIFA.
Hỗ trợ tìm kiếm Semantic và lọc theo Metadata.
Tim kiem san pham Canifa bang ProductSearchEngine core.
Wrapper nay giu compatibility voi schema cu va schema dual-lane moi.
"""
tool_start = time.time()
logger.info("🔧 data_retrieval_tool called with %d items", len(searches))
# Log search params cho debugging
for i, s in enumerate(searches):
logger.info(
"🔧 Search[%d]: desc=%r, name=%r, gender=%s, age=%s, color=%s, price=%s-%s, code=%s, mode=%s",
i,
(s.description[:80] + "...") if s.description and len(s.description) > 80 else s.description,
s.product_name,
s.gender_by_product,
s.age_by_product,
s.master_color,
s.price_min,
s.price_max,
s.magento_ref_code,
s.discovery_mode,
)
# Get DB Connection
db = get_db_connection()
if not db:
return json.dumps({"status": "error", "message": "Database connection failed"})
combined_results = []
all_filter_infos = []
tasks = [_execute_single_search(db, item) for item in searches]
results_list = await asyncio.gather(*tasks)
configurable = config.get("configurable", {}) if config else {}
db_source = configurable.get("db_source", "starrocks")
use_sqlite = db_source == "sqlite"
shared_user_insight = user_insight or configurable.get("user_insight")
for products, filter_info in results_list:
combined_results.extend(products)
if filter_info:
all_filter_infos.append(filter_info)
# Aggregate filter info from first result for simplicity in response
final_info = all_filter_infos[0] if all_filter_infos else {}
# Include search input so LLM knows its own reasoning
search_inputs = [
{
"description": item.description,
"gender_by_product": item.gender_by_product,
"age_by_product": item.age_by_product,
"discovery_mode": item.discovery_mode,
}
for item in searches
]
per_search_results: List[Dict[str, Any]] = []
all_products: List[Dict[str, Any]] = []
errors: List[str] = []
allowed_skus = sorted({str(p.get("sku") or "").strip() for p in combined_results if p.get("sku")})
output = {
"status": "success",
"message_for_ai": (
f"Quy tắc bắt buộc: `product_ids` CHỈ được lấy từ trường `sku`. "
f"Các giá trị `sku` hợp lệ trong lần search này là: {allowed_skus}. "
f"Tuyệt đối cấm dùng `sku_color` cho `product_ids`. "
f"`sku_color` chỉ là mã biến thể màu để tham chiếu nội bộ."
),
"search_input": search_inputs,
"results": combined_results,
"stock_skus": _resolve_stock_skus(searches, combined_results),
"filter_info": final_info,
}
total_ms = (time.time() - tool_start) * 1000
logger.info(
"🎁 Final result: %d products, total_ms=%.2f. Fallback used: %s",
len(combined_results),
total_ms,
final_info.get("fallback_used", False),
for item in searches:
literal_text = _build_literal_text(item)
logger.info("[Tool] Searching via ProductSearchEngine: %s", literal_text)
try:
result = await _run_product_search_engine(
item,
reasoning=reasoning,
user_insight=shared_user_insight,
)
except Exception as exc:
logger.exception("ProductSearchEngine failed; falling back to legacy engine: %s", exc)
result = await _run_legacy_engine(item, use_sqlite=use_sqlite)
if result.get("status") == "error":
errors.append(str(result.get("message", "unknown error")))
products = [_augment_product(p) for p in result.get("products", [])]
result["products"] = products
result["literal_text"] = result.get("literal_text") or literal_text
per_search_results.append(result)
all_products.extend(products)
products = _dedup_products(all_products)
has_success = any(r.get("status") == "success" for r in per_search_results)
has_products = bool(products)
status = "success" if has_success or has_products else ("error" if errors else "no_results")
payload = {
"status": status,
"count": len(products),
"products": products,
"results": per_search_results,
}
if errors:
payload["errors"] = errors
return json.dumps(output, ensure_ascii=False, default=str)
# Load dynamic docstring
dynamic_prompt = read_tool_prompt("data_retrieval_tool")
if dynamic_prompt:
data_retrieval_tool.__doc__ = dynamic_prompt
data_retrieval_tool.description = dynamic_prompt
return json.dumps(payload, ensure_ascii=False, default=str)
backend/agent/tools/get_tools.py
View file @ 94484304
"""
Tools Factory
Return all tools for the agent.
Tools Factory.
"""
from langchain_core.tools import Tool
from .brand_knowledge_tool import canifa_knowledge_search
from .data_retrieval_tool import data_retrieval_tool
from .promotion_canifa_tool import canifa_get_promotions
from .check_is_stock import check_is_stock
from .store_search_tool import canifa_store_search
def get_all_tools() -> list[Tool]:
"""Return toàn bộ list tools cho Agent"""
return [data_retrieval_tool, canifa_knowledge_search, canifa_get_promotions, check_is_stock, canifa_store_search]
from .brand_knowledge_tool import canifa_knowledge_search
from .check_is_stock import check_is_stock
from .data_retrieval_tool import data_retrieval_tool
from .promotion_canifa_tool import canifa_get_promotions
from .store_search_tool import canifa_store_search
return [
data_retrieval_tool,
canifa_knowledge_search,
canifa_get_promotions,
check_is_stock,
canifa_store_search,
]
backend/agent/tools/note/example.txt deleted 100644 → 0
View file @ f9b10ebe
{
"sku": "6KS25W005", ✅ Mã sản phẩm
"color_code": "SE008", ✅ Mã màu
"name": "Chân váy nữ dáng A",
"color": "Be/ Beige", ✅ Tên màu
"price": 399000,
"sale_price": 279300,
"url": "https://canifa.com/chan-vay-nu-6ks25w005?color=SE008&utm_source=chatbot&utm_medium=rsa&utm_campaign=ver1",
"thumbnail_image_url": "https://..."
}
\ No newline at end of file
backend/agent/tools/note/query.txt deleted 100644 → 0
View file @ f9b10ebe
-- [HYDE SEARCH] Full Executable Query
-- Original Params: ['women', 'unisex', 'adult', '%tím%', '%tím%', '%liền%']
WITH vector_matches AS (
SELECT /*+ SET_VAR(ann_params='{"ef_search":256}') */
internal_ref_code,
magento_ref_code,
product_color_code,
product_name,
master_color,
product_color_name,
product_image_url_thumbnail,
product_web_url,
sale_price,
original_price,
discount_amount,
ROUND(((original_price - sale_price) / original_price * 100), 0) as discount_percent,
age_by_product,
gender_by_product,
product_line_vn,
product_line_en,
description_text_full,
quantity_sold,
is_new_product,
approx_cosine_similarity(vector, [0.041171539574861526,0.03616797924041748,-0.04494577646255493,-0.028166595846414566,0.03670715540647507,-0.004836415871977806,0.005936336703598499,0.04503204673528671,0.06521882861852646,-0.014870496466755867,-0.050294410437345505,0.028080327436327934,-0.013997029513120651,-0.03474455326795578,0.02368064410984516,0.04800830036401749,-0.0194642823189497,0.002722842851653695,-0.030862480401992798,0.030129199847579002,0.013889194466173649,0.015280270017683506,0.012573602609336376,0.011214877478778362,0.018288876861333847,0.058317363262176514,0.03536999970674515,0.004480559378862381,0.02178274281322956,-0.03431321308016777,0.03142322227358818,-0.034528881311416626,0.024478627368807793,-0.05723900720477104,0.04498891159892082,0.021351400762796402,0.05051008239388466,0.01314513012766838,0.01698406971991062,-0.01670369692146778,0.025406010448932648,-0.00490111717954278,0.042249895632267,-0.009322366677224636,0.031078150495886803,0.029395919293165207,-0.0014369061682373285,0.0006557737942785025,0.021189648658037186,-0.003307849634438753,0.00994781218469143,-0.039942216128110886,-0.011171743273735046,0.043694887310266495,-0.02016521245241165,-0.00941941887140274,0.004868766292929649,0.011948158033192158,0.03028016909956932,-0.01120409369468689,0.007839630357921124,-0.04921605810523033,-0.027519583702087402,-0.0007036257302388549,-0.02976255863904953,-0.010810494422912598,-0.027929358184337616,-0.023184603080153465,0.0034857778809964657,-0.01089137140661478,0.04736128821969032,0.018051639199256897,-0.03317015618085861,0.02249445579946041,-0.006081914063543081,0.05473722890019417,-0.04268123582005501,-0.02639809623360634,-0.01051394734531641,-0.04076176509261131,-0.043177276849746704,0.030689943581819534,0.005345937795937061,0.03916580229997635,-0.03914423659443855,-0.02956845611333847,-0.027929358184337616,0.06478748470544815,0.014158782549202442,0.021394535899162292,0.0067612770944833755,-0.006513255648314953,0.04498891159892082,-0.005383680108934641,0.025406010448932648,-0.02506093680858612,0.0016364016337320209,-0.040222588926553726,-0.026354961097240448,0.043543919920921326,0.042789071798324585,-0.005688315257430077,0.035671938210725784,-0.013802926056087017,0.023249303922057152,-0.037720806896686554,0.013986245729029179,-0.020229913294315338,-0.05374514311552048,0.002678360790014267,-0.058878105133771896,-0.00401147548109293,-0.01755559630692005,0.042508698999881744,0.019496632739901543,0.025837352499365807,0.04268123582005501,0.03317015618085861,-0.023853180930018425,-0.03090561367571354,-0.012454983778297901,-0.005132962949573994,0.02288266271352768,-0.06056033819913864,-0.0351111926138401,0.04138721153140068,-0.011430548503994942,0.026031455025076866,-0.003046348923817277,-0.03493865579366684,-0.013846060261130333,0.04524771496653557,0.005273148883134127,-0.00780727993696928,0.010066431015729904,-0.017361493781208992,-0.051329631358385086,0.08014323562383652,-0.04593786224722862,0.008341064676642418,0.06780687719583511,-0.011053124442696571,0.05490976572036743,0.029870394617319107,0.011549167335033417,0.04503204673528671,-0.0019059899495914578,0.020801441743969917,-0.024953102692961693,0.05158843472599983,-0.0012994160642847419,0.004903812892735004,0.0035477832425385714,0.01065952517092228,-0.04106370359659195,-0.019755437970161438,-0.009139047004282475,0.0066642253659665585,-0.04641233757138252,0.018968239426612854,0.01618608832359314,-0.023723779246211052,0.011635434813797474,-0.027519583702087402,-0.032307472079992294,-0.0021391839254647493,0.018482981249690056,-0.014083297923207283,-0.05046694725751877,0.014514639042317867,0.005855460185557604,-0.011754054576158524,-0.008006775751709938,0.016779182478785515,-0.0002788554993458092,-0.03478768840432167,0.020758306607604027,0.015366538427770138,-0.0327603816986084,0.057454679161310196,0.04671427607536316,0.02197684533894062,-0.0028792039956897497,-0.06327778846025467,-0.08199800550937653,-0.0022483672946691513,-0.03972654789686203,0.01618608832359314,0.029439052566885948,-0.02368064410984516,-0.022580724209547043,0.010233575478196144,0.037569839507341385,-0.030107632279396057,0.025664815679192543,-0.05258052051067352,-0.012929460033774376,0.004774410743266344,0.030776211991906166,-0.011322712525725365,0.008702313527464867,-0.003612484550103545,-0.013899978250265121,-0.029201814904808998,0.021577855572104454,-0.03155262768268585,-0.06715986132621765,0.012627520598471165,0.05710960552096367,-0.02368064410984516,0.053615741431713104,0.005893202498555183,-0.001353333704173565,-0.02007894404232502,0.034485746175050735,-0.02700197324156761,0.04878471791744232,-0.004936163779348135,0.00414087763056159,0.02378848008811474,0.000368662120308727,0.009095912799239159,0.03066837601363659,0.029503753408789635,-0.01618608832359314,-0.003448035567998886,0.06392480432987213,-0.000943559396546334,-0.06940283626317978,-0.03470141813158989,-0.00039932780782692134,0.03834625333547592,0.017641864717006683,-0.028792040422558784,-0.021707257255911827,-0.018526114523410797,-0.019237827509641647,-0.006858328823000193,0.046110399067401886,0.05288245901465416,-0.027541151270270348,-0.0313153900206089,-0.011355062946677208,0.012012858875095844,0.0057584079913794994,-0.00911747943609953,-0.019518200308084488,-0.022300351411104202,0.03217807039618492,0.01614295318722725,0.062156300991773605,0.0035531751345843077,-0.06728926301002502,-0.021858226507902145,-0.020909275859594345,0.0073004537262022495,-0.04900038614869118,-0.006459338124841452,0.013943111523985863,-0.01452542282640934,0.04598099738359451,0.004445512779057026,-0.01018504984676838,-0.006200533360242844,0.00959195476025343,-0.016789965331554413,0.02790779061615467,0.07246536016464233,0.01089137140661478,0.04688681289553642,-0.0189898069947958,-0.026743169873952866,0.011635434813797474,0.04912979155778885,-0.01822417601943016,-0.012724572792649269,-0.010093390010297298,0.008572910912334919,0.039899084717035294,0.05370200797915459,-0.03858349099755287,-0.02169647440314293,0.043414514511823654,-0.031164418905973434,0.05288245901465416,0.04619666934013367,0.025190340355038643,-0.019572118297219276,0.0057314494624733925,-0.021610205993056297,-0.011829539202153683,0.02478056587278843,-0.04380272328853607,-0.03248000890016556,0.025039371103048325,0.007758754305541515,-0.026807870715856552,-0.011096258647739887,0.0034965614322572947,0.010400720871984959,0.038756027817726135,0.042228326201438904,0.003795804688706994,-0.007650918792933226,0.009759100154042244,0.0030409570317715406,0.0215994231402874,-0.049302324652671814,0.025880485773086548,-0.011754054576158524,0.02060733735561371,0.018957456573843956,-0.019345663487911224,-0.04498891159892082,-0.03157419338822365,-0.004636920522898436,-0.08324889838695526,0.01746932789683342,0.030409570783376694,0.003229669062420726,0.02577265165746212,0.023184603080153465,0.0598270557820797,0.005882418714463711,0.03198396787047386,0.007456814870238304,0.03213493525981903,-0.006577956955879927,0.03873446211218834,-0.032156504690647125,-0.06866955757141113,-0.025082504376769066,-0.011958940885961056,0.011452115140855312,0.011754054576158524,-0.015150868333876133,0.025902053341269493,0.010282101109623909,0.025147205218672752,0.040459826588630676,0.031725164502859116,-0.05184724181890488,-0.049302324652671814,0.007133308798074722,0.024241389706730843,0.010071822442114353,0.04043826088309288,-0.027994059026241302,-0.036858126521110535,0.02790779061615467,0.0030301737133413553,-0.0028899875469505787,-0.017027202993631363,0.030603675171732903,-0.01575474627315998,-0.03317015618085861,0.0654776319861412,-0.004844503477215767,-0.07759832590818405,0.05279619246721268,-0.016811532899737358,-0.07130073755979538,0.012023642659187317,-0.008966510184109211,-0.01065413374453783,-0.01978778839111328,0.05685080215334892,0.020855357870459557,0.010050255805253983,0.011408980935811996,0.018429063260555267,0.05736841261386871,0.008195486851036549,0.012454983778297901,-0.006815194617956877,0.060258399695158005,0.023982584476470947,-0.008669963106513023,0.02572951652109623,-0.0417754165828228,0.018148690462112427,-0.023055199533700943,-0.026139291003346443,0.018450630828738213,-0.0024856049567461014,-0.0013620953541249037,-0.003156880149617791,0.00020505567954387516,0.07440640032291412,-0.002927730092778802,-0.012994160875678062,-0.05046694725751877,0.029740992933511734,0.041840121150016785,-0.015431240200996399,-0.022904230281710625,0.09946733713150024,0.01288632582873106,-0.024047285318374634,-0.02991352789103985,0.024694297462701797,0.02183666080236435,-0.019280962646007538,0.010486988350749016,0.02948218770325184,0.012142261490225792,0.016595860943198204,-0.007931291125714779,0.012476551346480846,0.024435492232441902,0.0023575504310429096,-0.006081914063543081,0.037612974643707275,0.00975370779633522,0.022429754957556725,-0.021092595532536507,0.03554253280162811,0.0011949505424126983,-0.02098476141691208,0.009683615528047085,-0.013716657646000385,-0.01812712475657463,-0.0185153316706419,-0.011807971633970737,0.04050296172499657,0.03711692988872528,0.025406010448932648,-0.0016431412659585476,0.008540560491383076,-0.05870556831359863,0.0038443305529654026,-0.06797941029071808,0.04671427607536316,-0.017598731443285942,0.01267065480351448,-0.036771856248378754,-0.0002589733630884439,0.08005697280168533,0.0313153900206089,0.010983031243085861,0.02473743073642254,0.01072961837053299,-0.01497833151370287,-0.00010160113015444949,0.04102057218551636,0.00812000222504139,0.044730108231306076,-0.0051410505548119545,-0.03498179093003273,0.011667786166071892,-0.02462959662079811,-0.010336019098758698,-0.016811532899737358,-0.03459358215332031,0.012002075091004372,0.051286496222019196,-0.013511770404875278,-0.04086960107088089,-0.01836436241865158,-0.01338236778974533,0.058878105133771896,-0.007893548347055912,0.0313369557261467,0.010481596924364567,0.013511770404875278,0.0021728824358433485,-0.01841827854514122,-0.03976967930793762,0.025233473628759384,0.01359803881496191,-0.02277482859790325,-0.018774136900901794,0.03910110145807266,-0.012476551346480846,0.0038982483092695475,0.038518790155649185,0.043738022446632385,0.02387474849820137,-0.02331400476396084,-0.019539767876267433,-0.010848237201571465,0.03976967930793762,-0.022106248885393143,-0.034722987562417984,0.02007894404232502,0.0773826539516449,0.05542737618088722,0.028360700234770775,-0.04268123582005501,0.03437791392207146,-0.007456814870238304,0.015398888848721981,-0.005370201077312231,-0.008146961219608784,0.004822936374694109,-0.002352158771827817,0.01089137140661478,-0.0013000901089981198,0.005343242082744837,-0.050337545573711395,-0.006119656842201948,-0.033385828137397766,0.034572016447782516,-0.07863354682922363,0.03981281444430351,0.003000518772751093,0.03953244164586067,0.04869844764471054,0.019000589847564697,-0.002482909243553877,-0.026700034737586975,0.0006197163020260632,0.01695171929895878,0.0001299921568715945,-0.011473681777715683,0.007214185316115618,-0.018299659714102745,0.014007813297212124,-0.02311990037560463,0.003488473827019334,-0.008993469178676605,0.02122199907898903,0.024198254570364952,0.011355062946677208,-0.02344340644776821,-0.06349346041679382,0.006453946232795715,-0.025945186614990234,-0.0024883009027689695,-0.014396020211279392,-0.020531851798295975,0.004386203363537788,-0.026419663801789284,0.0022928493563085794,-0.0075808255933225155,0.009538037702441216,0.004092351999133825,-0.0005951164057478309,0.013608822599053383,-0.0019828227814286947,-0.002198493340983987,0.012476551346480846,-0.0012286491692066193,-0.01622922159731388,-0.03256627917289734,0.013878410682082176,-0.00048525910824537277,-0.0051140920259058475,-0.01695171929895878,0.012897108681499958,0.02885674126446247,-0.020855357870459557,-0.03834625333547592,-0.03603857755661011,-0.013749008066952229,0.002135140122845769,0.0005799520295113325,-0.029546888545155525,-0.014924413524568081,-0.030452705919742584,0.003933294676244259,0.015312621369957924,0.0058500682935118675,0.019162343814969063,-0.0016242701094597578,0.018310444429516792,-1.248952503374312e-05,0.043975260108709335,-0.016293922439217567,-0.03138009086251259,-0.029266515746712685,0.02249445579946041,0.007348979823291302,-0.04740442335605621,0.008109219372272491,-0.027735253795981407,0.023961016908288002,-0.00865917932242155,-0.0055966549552977085,-0.02743331529200077,-0.02805875986814499,-0.0065294308587908745,0.00671814288944006,-0.022839529439806938,-0.03431321308016777,0.014180350117385387,0.0379580482840538,-0.03364463150501251,-0.041840121150016785,0.012314798310399055,0.006825978402048349,-0.02093084342777729,0.00037472788244485855,0.03155262768268585,0.02572951652109623,-0.013878410682082176,-0.028468534350395203,-0.013447069562971592,-0.016488026827573776,-0.060646604746580124,0.02639809623360634,-0.01723209023475647,-0.03437791392207146,-0.00028357328847050667,-0.004653095733374357,0.008707704953849316,0.0015703524695709348,-0.007408289238810539,-0.00582850119099021,-0.020542636513710022,0.011026165448129177,0.04395369067788124,0.01674683205783367,-0.007316628936678171,0.01444993820041418,-0.0028145029209554195,-0.03301918879151344,0.03651305288076401,-0.02383161522448063,-0.02639809623360634,-0.029827259480953217,-0.00996398739516735,-0.012616736814379692,-0.021049462258815765,-0.0031946224626153708,0.02620399184525013,-0.01689780130982399,-0.06474435329437256,-0.016919367015361786,-0.023098334670066833,0.0072034019976854324,-0.030625242739915848,0.03478768840432167,-0.00797442439943552,0.01751246303319931,-0.015291053801774979,-0.022753261029720306,-0.035585667937994,0.024564895778894424,0.00011617575364653021,-0.012487334199249744,0.002071786904707551,0.04542025178670883,-0.018731001764535904,0.03575820475816727,0.011117825284600258,-0.026743169873952866,-0.01521556917577982,-0.01585179753601551,0.024133553728461266,0.029266515746712685,0.0010271318024024367,0.03435634449124336,0.018677083775401115,0.020855357870459557,0.025902053341269493,-0.04727502167224884,-0.0074190725572407246,-0.030172333121299744,0.015938065946102142,-0.027562716975808144,0.018687868490815163,0.01274613942950964,-0.025751084089279175,0.004275672137737274,0.01314513012766838,0.013964679092168808,0.018008505925536156,-0.020639687776565552,-0.013986245729029179,0.0021715345792472363,0.041602883487939835,0.023141467943787575,0.0006648724083788693,-0.025362877175211906,-0.007753362413495779,-0.00966204795986414,0.01894667185842991,0.034334778785705566,-0.017728133127093315,-0.008928767405450344,0.0189790241420269,0.022278785705566406,0.011258011683821678,0.030409570783376694,-0.006464730016887188,-0.005122179631143808,0.04039512574672699,-0.03295448422431946,0.010508555918931961,0.008389591239392757,0.004661183338612318,0.012066776864230633,0.010524731129407883,0.03390343859791756,-0.006653441581875086,0.009451769292354584,0.04524771496653557,0.002025956753641367,-0.003728407435119152,-0.029935095459222794,0.004857982974499464,-0.030452705919742584,-0.021707257255911827,-0.04265966638922691,0.009796842001378536,-0.03241530805826187,-0.03301918879151344,0.0006914942641742527,-0.0037877170834690332,0.001671448117122054,-0.016434108838438988,-0.01203442644327879,0.02605302259325981,0.017124254256486893,0.0061088730581104755,-0.012713789008557796,-0.017210522666573524,-0.012853974476456642,0.009101304225623608,-0.018332011997699738,0.005521170329302549,-0.029697857797145844,-0.0008593130041845143,-0.05680766701698303,0.035434700548648834,0.04818083718419075,0.019960325211286545,-0.0023723780177533627,0.003121833549812436,0.020618120208382607,0.028511669486761093,-0.017005635425448418,0.026980407536029816,-0.023723779246211052,-0.026096157729625702,-0.006896071135997772,0.02620399184525013,-0.00048290021368302405,-0.017771266400814056,0.003919815178960562,-0.003499257378280163,0.03420537710189819,0.0057584079913794994,-0.0204132329672575,0.043177276849746704,-0.013759791851043701,-0.00582850119099021,-0.007920507341623306,0.02292579784989357,0.01808398962020874,0.004779802169650793,-0.04718875139951706,0.035715069621801376,0.008767014369368553,-0.0018925105687230825,0.009414026513695717,0.013544120825827122,0.06828135251998901,-0.0006187053513713181,-0.02084457501769066,0.039230503141880035,-0.004671967122703791,-0.009079737588763237,-0.035585667937994,0.010869803838431835,-0.02648436464369297,0.007478381972759962,-0.05715274065732956,-0.0011700136819854379,0.021944494917988777,-0.009980162605643272,-0.03161732852458954,0.004162444733083248,-0.014406803995370865,-0.0007177790976129472,-0.03981281444430351,-0.0028360700234770775,0.015743961557745934,0.015571425668895245,0.0074406396597623825,0.0056182220578193665,-0.0075376913882792,-0.025082504376769066,0.02795092575252056,-0.03345052897930145,0.007214185316115618,0.008702313527464867,0.004429337568581104,-0.027692120522260666,0.014352886006236076,0.011559950187802315,0.016078252345323563,-0.03271724656224251,0.0389285646378994,0.009888502769172192,-0.012347148731350899,-0.008405766449868679,0.009300800040364265,-0.04636920616030693,-0.026549065485596657,0.016919367015361786,-0.05111395940184593,0.01709190383553505,0.02572951652109623,0.008319498039782047,-0.019830921664834023,0.04332824796438217,0.025362877175211906,-0.03301918879151344,-0.011430548503994942,0.005348633974790573,-0.0043996828608214855,0.03155262768268585,0.05715274065732956,0.002981647616252303,0.011710920371115208,-0.014352886006236076,-0.02833913266658783,0.012832407839596272,0.07293983548879623,-0.006626483052968979,-0.012314798310399055,0.017307575792074203,-0.009570388123393059,0.013350017368793488,0.04076176509261131,0.02296893112361431,0.014277401380240917,-0.0061088730581104755,-0.029978230595588684,0.016078252345323563,0.004092351999133825,-0.036901261657476425,-0.04054609686136246,0.0011774273589253426,-0.00262309517711401,-0.02083379216492176,-0.005227318964898586,0.011710920371115208,0.040373560041189194,0.01737227663397789,0.00539985578507185,-0.023896316066384315,-0.015129300765693188,0.016638996079564095,0.012109911069273949,-0.017695782706141472,0.009338541887700558,0.06099167838692665,-0.002104137558490038,0.002100093523040414,0.014115648344159126,0.01798693835735321,-0.04283220320940018,0.010848237201571465,0.011969724670052528,0.0168546661734581,0.02192292921245098,0.014126432128250599,-0.02482369914650917,0.010389937087893486,-0.023098334670066833,0.01717817224562168,-0.007198010105639696,-0.035866040736436844,0.003383334493264556,0.02093084342777729,-0.005488819908350706,0.061207350343465805,-0.01671447977423668,0.00505478261038661,0.0024977365974336863,0.014697959646582603,0.01646645925939083,0.026937272399663925,-0.018526114523410797,0.050294410437345505,0.013242182321846485,0.036901261657476425,0.03390343859791756,0.01870943419635296,0.025880485773086548,0.013943111523985863,-0.014622475020587444,0.0013951199362054467,-0.01430975180119276,0.0379796139895916,-0.02596675418317318,0.013781358487904072,0.009775275364518166,0.006103481166064739,0.007985208183526993,0.014945981092751026,0.00431880634278059,0.010546297766268253,0.008179311640560627,0.005882418714463711,0.020909275859594345,-0.02877047471702099,0.008163136430084705,0.030970314517617226,-0.0051572262309491634,0.006394636817276478,-0.013468636199831963,0.042551834136247635,0.011408980935811996,0.01998189277946949,-0.013759791851043701,0.018332011997699738,0.007478381972759962,-0.00366640230640769,-0.014719526283442974,-0.026980407536029816,0.008864066563546658,0.02013286203145981,0.02378848008811474,-0.025147205218672752,0.05025127902626991,-0.04291847348213196,0.02810189500451088,-0.00935471709817648,-0.02672160230576992,-0.030258601531386375,-0.028037194162607193,-0.035003356635570526,0.0038793771527707577,-0.02710980921983719,0.015463590621948242,-0.0017536724917590618,-0.02136218547821045,0.020014243200421333,-0.021146513521671295,0.013716657646000385,0.0005428836448118091,-0.0050574783235788345,0.011958940885961056,-0.00041482914821244776,0.002569177420809865,-0.034291643649339676,-0.01322061475366354,-0.004844503477215767,-0.037763942033052444,0.0035396956373006105,0.015733178704977036,0.041753850877285004,0.013943111523985863,-0.01970151998102665,0.02203076332807541,0.00412739859893918,0.010567865334451199,-0.00925766583532095,0.022666992619633675,0.0031056583393365145,-0.01780361868441105,0.0016215741634368896,0.012476551346480846,-0.013608822599053383,-0.008486642502248287,-0.0029223382007330656,0.0074136811308562756,0.012066776864230633,0.0018399407854303718,-0.008777798153460026,0.030452705919742584,-0.017242873087525368,0.030258601531386375,-0.03698752820491791,0.06634031236171722,0.02572951652109623,-0.017922237515449524,0.025556979700922966,0.00579075887799263,0.01812712475657463,0.0035181285347789526,0.03483081981539726,0.038712892681360245,-0.007990600541234016,-0.0069014630280435085,-0.05158843472599983,0.03534843027591705,0.05193350836634636,0.05063948407769203,0.022516023367643356,-0.009640481323003769,0.009219923056662083,-0.007009298540651798,-0.008400374092161655,0.05197664350271225,0.009500294923782349,-0.041559748351573944,0.011026165448129177,-0.0005910725449211895,0.011171743273735046,-0.010540906339883804,0.00828175526112318,-0.0033779426012188196,0.03813058137893677,-0.016293922439217567,0.02616085857152939,0.006276017986238003,-0.029223382472991943,0.00487685389816761,0.025082504376769066,-0.032587844878435135,0.03886386379599571,0.03584447503089905,0.009365500882267952,0.0038497222121804953,-0.021998412907123566,-0.019011374562978745,-0.03127225488424301,0.01203442644327879,0.028317565098404884,0.0033860302064567804,0.02065047062933445,-0.01474109385162592,0.03248000890016556,0.04675741121172905,-0.0006823956500738859,-0.0194642823189497,0.012282447889447212,0.03327799215912819,-0.013554904609918594,0.00575301656499505,0.03817371651530266,0.030862480401992798,-0.013684307225048542,-0.03502492606639862,0.008998860605061054,0.009505687281489372,-0.0012926763156428933,-0.05301186442375183,0.0018803790444508195,-0.0225375909358263,0.011214877478778362,-0.0036017009988427162,0.02084457501769066,-0.003798500634729862,0.00965665653347969,-0.006631874479353428,-0.009333150461316109,0.01614295318722725,-0.003022085875272751,-0.018968239426612854,-0.023184603080153465,-0.005256973672658205,0.00787737313657999,0.0025085201486945152,0.0010365673806518316,-0.02842540107667446,-0.005947120022028685,-0.0021068332716822624,0.03360149636864662,-0.00644316291436553,0.00715487590059638,0.018332011997699738,-0.017296791076660156,0.031444791704416275,0.0004724536556750536,-0.01580866426229477,-0.015603776089847088,0.011430548503994942,0.03573663905262947,-0.0004936837358400226,0.019809355959296227,-0.006755885202437639,0.02782152220606804,-0.016358623281121254,-0.01884962059557438,0.022063113749027252,-0.009311582893133163,0.0006699271616525948,0.012800057418644428,0.036081712692976,-0.016822315752506256,-0.015970416367053986,-0.013662739656865597,-0.0012171915732324123,0.03498179093003273,-0.027282346040010452,0.004281063564121723,0.019712302833795547,0.005065565928816795,-0.015409672632813454,-0.01336080115288496,0.010012513026595116,-0.01652037724852562,0.010632566176354885,-0.010907546617090702,-0.011937374249100685,-0.007516124751418829,0.008195486851036549,0.0185153316706419,0.021815093234181404,-0.03841095417737961,0.02335713803768158,0.028943009674549103,0.03854035586118698,0.01790066994726658,0.0031811431981623173,0.010610999539494514,-0.012810840271413326,0.017879102379083633,-0.047533825039863586,0.011376630514860153,-0.02695883996784687,0.019302530214190483,-0.029978230595588684,0.03118598647415638,0.019669169560074806,0.014568557031452656,-0.0088964169844985,-0.0003381649439688772,0.0027794563211500645,0.0107997115701437,0.019572118297219276,0.018105557188391685,0.012552035972476006,0.010411503724753857,-0.026570633053779602,-0.003925207071006298,-0.0159057155251503,-0.0016296618850901723,-0.014191132970154285,0.0029412093572318554,-0.026225559413433075,-0.017480112612247467,0.0033132415264844894,-0.012649087235331535,0.019162343814969063,0.012422633357346058,-0.02430609054863453,0.02340027317404747,-0.028209729120135307,-0.004105831496417522,0.002765977056697011,-0.022192517295479774,-0.015398888848721981,-0.017102688550949097,0.04710248485207558,-0.0044050742872059345,0.018569249659776688,-0.021966062486171722,-5.214850534684956e-05,-0.005580479744821787,0.037483569234609604,0.06478748470544815,-0.027605852112174034,-0.006707359571009874,-0.008643004111945629,-0.013705873861908913,0.0038658976554870605,-0.02073673903942108,0.017825184389948845,0.007122525479644537,0.009300800040364265,0.009791450574994087,0.017027202993631363,0.005774583667516708,0.032156504690647125,0.007149484474211931,-0.010837453417479992,-0.008324889466166496,-0.02581578493118286,-0.0019545159302651882,-0.014967547729611397,-0.005526562221348286,0.004604569636285305,-0.007942073978483677,-0.04774949699640274,0.012897108681499958,-0.031099718064069748,-0.001965299481526017,0.006993122864514589,0.03640521690249443,0.044730108231306076,0.017534028738737106,0.03838938847184181,-0.018116340041160583,-0.027260778471827507,-0.013964679092168808,-0.04104213789105415,0.01989562436938286,-0.030862480401992798,0.02510407194495201,-0.017124254256486893,-0.005877027288079262,-0.008443508297204971,0.022429754957556725,-0.0014085994334891438,0.015927283093333244,-0.003814675845205784,-0.013608822599053383,-0.02340027317404747,0.060172129422426224,0.0010985727421939373,-0.021825876086950302,0.013274532742798328,-0.002113573020324111,-0.010481596924364567,0.00987232755869627,0.01823495887219906,0.01879570260643959,0.0032269731163978577,-0.032674115151166916,0.01072961837053299,0.016035117208957672,-0.03439947962760925,0.018051639199256897,0.004097743425518274,-0.004423945676535368,0.012853974476456642,0.006567173171788454,0.055944982916116714,-0.005027823615819216,0.0012488682987168431,0.016908584162592888,0.00286033283919096,0.006939205341041088,-0.022278785705566406,0.00435654865577817,-0.017738915979862213,0.02021913044154644,-0.02629026025533676,-0.015787096694111824,-0.037009093910455704,-0.02596675418317318,0.00828175526112318,0.012703005224466324,0.011937374249100685,-0.018159475177526474,-0.04969053342938423,-0.028684206306934357,0.016692914068698883,-0.03603857755661011,-0.013069645501673222,-0.0009368197061121464,0.0065240394324064255,-0.0026689250953495502,-0.0102012250572443,-0.02268856018781662,-0.012713789008557796,-0.03877759352326393,0.02790779061615467,-0.020801441743969917,0.0002916609519161284,-0.0022133206948637962,0.025902053341269493,-0.006971555761992931,0.01989562436938286,0.007068607956171036,-0.014546990394592285,-0.00897190161049366,0.008545951917767525,-0.00524619035422802,-0.03698752820491791,-0.006626483052968979,0.02288266271352768,0.011408980935811996,-0.021588638424873352,-0.028554802760481834,0.0012016902910545468,-0.027778388932347298,-0.024758998304605484,-0.01322061475366354,-0.03722476586699486,0.025751084089279175,-0.05434902012348175,-0.0017078424571081996,0.023723779246211052,-0.0389069989323616,0.01298337709158659,0.016304707154631615,0.05201977863907814,0.015096950344741344,-0.011732487007975578,-0.0022483672946691513,-0.021135730668902397,-0.008324889466166496,-0.00503051932901144,0.01927017793059349,-0.00012813873763661832,-0.008885633200407028,-0.038475655019283295,-0.016358623281121254,-0.00233867927454412,0.0055912635289132595,0.011333496309816837,0.019863273948431015,0.012023642659187317,0.0016903192736208439,0.011872673407196999,0.004577611107379198,0.01559299323707819,-0.007920507341623306,-0.020639687776565552,0.04593786224722862,-0.006114264950156212,0.02976255863904953,-0.01932409591972828,0.031358521431684494,0.003663706360384822,0.0035693503450602293,-0.011710920371115208,0.024478627368807793,0.00249234470538795,0.011667786166071892,0.004283759742975235,-0.01545280683785677,-0.004014171194285154,0.015743961557745934,0.012595170177519321,-0.015043032355606556,0.022365054115653038,-0.027303913608193398,-0.0230120662599802,-0.02838226594030857,-0.0032242771703749895,0.016531160101294518,0.02458646148443222,0.006141223944723606,0.020909275859594345,-0.006847545504570007,0.004990081302821636,0.004874158184975386,-0.0025840047746896744,-0.008767014369368553,0.0005068262107670307,-0.013403935357928276,-0.03090561367571354,-0.02833913266658783,-0.02877047471702099,0.0023265478666871786,-0.017544813454151154,0.028209729120135307,0.022710125893354416,0.03297605365514755,0.009721357375383377,0.004787889774888754,0.005558912642300129,0.016887016594409943,0.02486683428287506,0.005893202498555183,0.007176443003118038,0.006825978402048349,0.026311827823519707,-0.011764837428927422,-0.0067828441970050335,0.037052229046821594,0.02193371206521988,0.010168874636292458,-0.011829539202153683,0.013608822599053383,0.005205751862376928,-0.012767706997692585,-0.017868319526314735,0.003067916026338935,-0.009300800040364265,-0.0026621853467077017,-0.02277482859790325,-0.04580846056342125,0.01955055072903633,0.043457649648189545,0.030625242739915848,-0.042228326201438904,-0.011225661262869835,0.004585698712617159,0.028921443969011307,-0.02771368809044361,-0.0026635334361344576,-0.025600114837288857,0.02060733735561371,-0.009333150461316109,-0.04774949699640274,0.023745346814393997,-0.036534618586301804,-0.01476266048848629,0.05676453188061714,0.02122199907898903,0.02140531875193119,-0.03166045993566513,-0.0037796292454004288,-0.0020731347613036633,0.009705182164907455,-0.001767151989042759,-0.04229302704334259,0.0016390974633395672,-0.026851003989577293,-0.04662800952792168,0.058834973722696304,-0.04550652205944061,-0.02767055295407772,-0.010077213868498802,-0.008179311640560627,-0.006400028709322214,0.028403833508491516,0.015830229967832565,0.012325581163167953,0.0332132913172245,0.0010129783768206835,-0.021901361644268036,-0.00733280461281538,0.004388899076730013,-0.022278785705566406,0.01652037724852562,-0.006739709991961718,-0.010411503724753857,0.00041988393059000373,0.026937272399663925,0.013468636199831963,0.019755437970161438,0.009624306112527847,-0.03118598647415638,-0.024931535124778748,0.05443529039621353,0.01112860906869173,0.02596675418317318,0.004693534225225449,-0.01414799876511097,0.004510214086622,-0.02980569377541542,0.03138009086251259,-0.005480732303112745,-0.006044171750545502,0.027692120522260666,-0.00828175526112318,-0.0389285646378994,0.04947486147284508,-0.0025745693128556013,0.0031838389113545418,-0.0008660527528263628,-0.024715865030884743,-0.0014180350117385387,0.03312702104449272,-0.008783189579844475,-0.002701275749132037,-0.02311990037560463,0.006281409878283739,-0.0008903156849555671,-0.010179657489061356,0.028986144810914993,0.056548863649368286,-0.0168546661734581,-0.01267065480351448,-0.032587844878435135,-0.016488026827573776,0.018644733354449272,0.022947363555431366,4.0480383177055046e-05,0.024457059800624847,-0.00942481029778719,0.03198396787047386,0.028446968644857407,-0.0021863619331270456,0.011861889623105526,0.02292579784989357,0.0008384199463762343,0.027303913608193398,0.014342103153467178,0.0002136488037649542,-0.01312356349080801,-0.00334020028822124,0.012541252188384533,0.026807870715856552,0.0038416346069425344,-0.0045668273232877254,0.023227736353874207,-0.0006729600136168301,0.026786303147673607,-0.010260534472763538,-0.03222120553255081,0.003162271808832884,-0.021114163100719452,0.010621783323585987,0.008944942615926266,0.002895379438996315,0.002255107043311,0.02592362090945244,-0.02913711406290531,-0.0037769335322082043,0.0015784400748088956,-0.025039371103048325,0.009144438430666924,0.004000691697001457,0.017404627054929733,-0.022861097007989883,0.006324543617665768,0.028490101918578148,0.031207552179694176,0.026570633053779602,0.011355062946677208,-0.004928075708448887,0.00788276456296444,-0.009478728286921978,-0.02691570483148098,0.0006527408841066062,-0.011171743273735046,0.015398888848721981,-0.024845266714692116,0.019097642973065376,0.049345459789037704,0.014881279319524765,0.03944617509841919,-0.008928767405450344,-0.03979124873876572,-0.009505687281489372,-0.012217746116220951,0.02434922382235527,0.021631773561239243,0.0075754341669380665,0.037375736981630325,0.0027605851646512747,-0.0024357312358915806,0.025643248111009598,0.008292539045214653,0.0159057155251503,-0.023270869627594948,-0.051329631358385086,0.02700197324156761,-0.03379560261964798,0.017264440655708313,0.01812712475657463,0.005418726708739996,-0.025664815679192543,0.002643314190208912,0.013403935357928276,-0.022796394303441048,-0.02534130960702896,0.010265925899147987,-0.008578303270041943,0.011926590465009212,0.012314798310399055,0.0019113817252218723,0.016337057575583458,-0.026570633053779602]) as similarity_score
FROM shared_source.magento_product_dimension_with_text_embedding
ORDER BY similarity_score DESC
LIMIT 200
),
filtered_matches AS (
SELECT * FROM vector_matches
WHERE gender_by_product IN ('women', 'unisex') AND age_by_product = 'adult' AND (LOWER(master_color) LIKE '%tím%' OR LOWER(product_color_name) LIKE '%tím%') AND LOWER(product_name) LIKE '%liền%'
ORDER BY similarity_score DESC
LIMIT 150
)
SELECT
internal_ref_code,
MAX_BY(magento_ref_code, similarity_score) as magento_ref_code,
product_color_code,
MAX_BY(product_name, similarity_score) as product_name,
MAX_BY(master_color, similarity_score) as master_color,
MAX_BY(product_image_url_thumbnail, similarity_score) as product_image_url_thumbnail,
MAX_BY(product_web_url, similarity_score) as product_web_url,
MAX_BY(sale_price, similarity_score) as sale_price,
MAX_BY(original_price, similarity_score) as original_price,
MAX_BY(discount_amount, similarity_score) as discount_amount,
MAX_BY(discount_percent, similarity_score) as discount_percent,
MAX_BY(description_text_full, similarity_score) as description_text_full,
MAX_BY(gender_by_product, similarity_score) as gender_by_product,
MAX_BY(age_by_product, similarity_score) as age_by_product,
MAX_BY(product_line_vn, similarity_score) as product_line_vn,
MAX_BY(quantity_sold, similarity_score) as quantity_sold,
MAX(similarity_score) as max_score
FROM filtered_matches
GROUP BY product_color_code, internal_ref_code
ORDER BY max_score DESC
LIMIT 80
\ No newline at end of file
backend/agent/tools/promotion_canifa_tool.py
View file @ 94484304
......@@ -12,7 +12,7 @@ logger = logging.getLogger(__name__)
class PromotionInput(BaseModel):
check_date: str | None = Field(
...,
default=None,
description="Ngày cần kiểm tra khuyến mãi (định dạng YYYY-MM-DD). Nếu muốn kiểm tra ngày hiện tại thì truyền null."
)
model_config = {"extra": "forbid"}
......@@ -22,19 +22,12 @@ class PromotionInput(BaseModel):
async def canifa_get_promotions(check_date: str = None) -> str:
"""
Tra cứu danh sách các chương trình khuyến mãi (CTKM) đang diễn ra theo ngày.
Sử dụng tool này khi khách hàng hỏi về:
- "Hôm nay có khuyến mãi gì không?"
- "Đang có chương trình gì hot?"
- "Ngày mai có giảm giá không?"
- "Danh sách mã giảm giá hiện tại."
Trả về: Tên chương trình, mô tả chi tiết, thời gian áp dụng.
"""
target_date = check_date
if not target_date:
target_date = datetime.now().strftime("%Y-%m-%d")
logger.info(f"🎁 [Promotion Search] Checking for date: {target_date}")
logger.info(f"🎁 [Promotion Search] Checking for date: {target_date} (StarRocks)")
try:
sql = f"""
......@@ -64,9 +57,8 @@ async def canifa_get_promotions(check_date: str = None) -> str:
results = await sr.execute_query_async(sql)
if not results:
return f"Hiện tại (ngày {target_date}) không có chương trình khuyến mãi nào đang diễn ra trên hệ thống."
return f"Hiện tại (ngày {target_date}) không có chương trình khuyến mãi nào đang diễn ra."
# Map channel values to Vietnamese labels
channel_labels = {
"only_online": "Online (Web/App)",
"both": "Online + Cửa hàng",
......@@ -83,8 +75,6 @@ async def canifa_get_promotions(check_date: str = None) -> str:
t_date = res.get("to_date", "")
channel = res.get("applied_channel", "unknown")
channel_label = channel_labels.get(channel, channel)
# Use description_full if available and longer, else description
content = desc_full if desc_full and len(str(desc_full)) > len(str(desc)) else desc
lines.append(
......@@ -95,21 +85,10 @@ async def canifa_get_promotions(check_date: str = None) -> str:
f"Thời gian: {f_date} đến {t_date}\n"
)
lines.append(
"LƯU Ý QUAN TRỌNG VỀ KÊNH ÁP DỤNG:\n"
"- Mặc định: ƯU TIÊN trình bày các CTKM 'Online (Web/App)' và 'Online + Cửa hàng' TRƯỚC.\n"
"- Chỉ trình bày CTKM 'Chỉ tại cửa hàng' khi khách hỏi CỤ THỂ về ưu đãi tại cửa hàng/offline.\n"
"- Trình bày NỘI DUNG ưu đãi chi tiết cho khách. "
"Nếu nội dung chỉ ghi địa điểm áp dụng mà KHÔNG ghi cụ thể giảm bao nhiêu %, "
"hãy trình bày tên chương trình + thời gian + nội dung có sẵn, "
"rồi hướng dẫn khách liên hệ hotline 1800 6061 hoặc vào canifa.com để xem chi tiết ưu đãi."
)
return "\n".join(lines)
except Exception as e:
logger.error(f"❌ Error in canifa_get_promotions: {e}")
return "Xin lỗi, tôi không thể lấy danh sách khuyến mãi lúc này."
# Load dynamic docstring
canifa_get_promotions.__doc__ = read_tool_prompt("promotion_canifa_tool") or canifa_get_promotions.__doc__
backend/agent/tools/save.py deleted 100644 → 0
View file @ f9b10ebe
import logging
from common.embedding_service import create_embedding_async
logger = logging.getLogger(__name__)
def _escape(val: str) -> str:
"""Thoát dấu nháy đơn để tránh SQL Injection cơ bản."""
return val.replace("'", "''")
def _get_where_clauses(params) -> list[str]:
"""
Xây dựng WHERE clauses theo thứ tự ưu tiên dựa trên selectivity thực tế
FILTER PRIORITY (Based on Canifa catalog analysis):
🔥 TIER 1 (99% selectivity):
1. SKU Code → 1-5 records
🎯 TIER 2 (50-70% selectivity):
2. Gender → Splits catalog in half
3. Age → Kids vs Adults split
4. Product Category → 10-15 categories
💎 TIER 3 (30-50% selectivity):
5. Material Group → Knit vs Woven (2 groups)
6. Price Range → Numeric filtering
🎨 TIER 4 (10-30% selectivity):
7. Season → 4 seasons
8. Style/Fitting → Multiple options
⚠️ TIER 5 (<10% selectivity):
9. Form details → Granular attributes
10. Color → LOWEST selectivity (many SKUs share colors)
Early return: If SKU exists, skip low-selectivity filters
"""
clauses = []
# 🔥 TIER 1: SKU/Product Code (Unique identifier)
# Selectivity: ~99% → 1 SKU = 1 style (3-5 colors max)
sku_clause = _get_sku_clause(params)
if sku_clause:
clauses.append(sku_clause)
# Early return optimization: SKU đã xác định product rõ ràng
# CHỈ GIỮ LẠI price filter (nếu có) để verify budget constraint
# BỎ QUA: gender, color, style, fitting... vì SKU đã unique
price_clauses = _get_price_clauses(params)
if price_clauses:
clauses.extend(price_clauses)
return clauses # ⚡ STOP - Không thêm filter khác!
# 🎯 TIER 2: High-level categorization (50-70% reduction)
# Gender + Age + Category có selectivity cao nhất trong non-SKU filters
clauses.extend(_get_high_selectivity_clauses(params))
# 💎 TIER 3: Material & Price (30-50% reduction)
material_clause = _get_material_clause(params)
if material_clause:
clauses.append(material_clause)
clauses.extend(_get_price_clauses(params))
# 🎨 TIER 4: Attributes (10-30% reduction)
clauses.extend(_get_attribute_clauses(params))
# ⚠️ TIER 5: Granular details & Color (LAST - lowest selectivity)
clauses.extend(_get_form_detail_clauses(params))
color_clause = _get_color_clause(params)
if color_clause:
clauses.append(color_clause) # Color ALWAYS LAST!
return clauses
def _get_sku_clause(params) -> str | None:
"""
TIER 1: SKU/Product Code (Highest selectivity - 99%)
1 SKU code = 1 product style (may have 3-5 color variants)
WHY SKU is always priority #1:
- 1 code = 1 unique product design
- Adding other filters (color, style, gender) is redundant
- Only price filter may be kept for budget validation
Example queries:
- "Mã 6OT25W010" → Only SKU needed
- "Mã 6OT25W010 màu xám" → Only SKU (color is for display/selection, not filtering)
- "Mã 6OT25W010 dưới 500k" → SKU + price (validate budget)
"""
m_code = getattr(params, "magento_ref_code", None)
if m_code:
m = _escape(m_code)
return f"(magento_ref_code = '{m}' OR internal_ref_code = '{m}')"
return None
def _get_color_clause(params) -> str | None:
"""
TIER 5: Color (LOWEST selectivity - 5-10%)
Multiple SKUs share the same color (e.g., 50+ gray products)
ALWAYS filter color LAST after other constraints
"""
color = getattr(params, "master_color", None)
if color:
c = _escape(color).lower()
return f"(LOWER(master_color) LIKE '%{c}%' OR LOWER(product_color_name) LIKE '%{c}%')"
return None
def _get_high_selectivity_clauses(params) -> list[str]:
"""
TIER 2: High-level categorization (50-70% reduction per filter)
Order: Gender → Age → Product Category
"""
clauses = []
# Gender: Male/Female/Unisex split (50-70% reduction)
gender = getattr(params, "gender_by_product", None)
if gender:
clauses.append(f"gender_by_product = '{_escape(gender)}'")
# Age: Kids/Adults split (50% reduction of remaining)
age = getattr(params, "age_by_product", None)
if age:
clauses.append(f"age_by_product = '{_escape(age)}'")
# Product Category: Váy/Áo/Quần... (30-50% reduction)
product_line = getattr(params, "product_line_vn", None)
if product_line:
p = _escape(product_line).lower()
clauses.append(f"LOWER(product_line_vn) LIKE '%{p}%'")
return clauses
def _get_material_clause(params) -> str | None:
"""TIER 3: Material Group - Knit vs Woven (50% split)"""
material = getattr(params, "material_group", None)
if material:
m = _escape(material).lower()
return f"LOWER(material_group) LIKE '%{m}%'"
return None
def _get_price_clauses(params) -> list[str]:
"""TIER 3: Price Range - Numeric filtering (30-40% reduction)"""
clauses = []
p_min = getattr(params, "price_min", None)
if p_min is not None:
clauses.append(f"sale_price >= {p_min}")
p_max = getattr(params, "price_max", None)
if p_max is not None:
clauses.append(f"sale_price <= {p_max}")
return clauses
def _get_attribute_clauses(params) -> list[str]:
"""
TIER 4: Attributes (10-30% reduction)
Season, Style, Fitting
"""
clauses = []
# Season: 4 seasons (~25% each)
season = getattr(params, "season", None)
if season:
s = _escape(season).lower()
clauses.append(f"LOWER(season) LIKE '%{s}%'")
# Style: Basic/Feminine/Sporty... (~15-20% reduction)
style = getattr(params, "style", None)
if style:
st = _escape(style).lower()
clauses.append(f"LOWER(style) LIKE '%{st}%'")
# Fitting: Regular/Slim/Loose (~15% reduction)
fitting = getattr(params, "fitting", None)
if fitting:
f = _escape(fitting).lower()
clauses.append(f"LOWER(fitting) LIKE '%{f}%'")
# Size Scale: S, M, L, 29, 30... (Specific filtering)
size = getattr(params, "size_scale", None)
if size:
sz = _escape(size).lower()
clauses.append(f"LOWER(size_scale) LIKE '%{sz}%'")
return clauses
def _get_form_detail_clauses(params) -> list[str]:
"""
TIER 5: Granular form details (<10% reduction each)
Neckline, Sleeve type
"""
clauses = []
form_fields = [
("form_neckline", "form_neckline"),
("form_sleeve", "form_sleeve"),
]
for param_name, col_name in form_fields:
val = getattr(params, param_name, None)
if val:
v = _escape(val).lower()
clauses.append(f"LOWER({col_name}) LIKE '%{v}%'")
return clauses
async def build_starrocks_query(params, query_vector: list[float] | None = None) -> str:
"""
Build SQL Hybrid tối ưu với Filter Priority:
1. Pre-filtering theo độ ưu tiên (SKU → Exact → Price → Partial)
2. Vector Search (HNSW Index) - Semantic understanding
3. Flexible Keyword Search (OR + Scoring) - Fuzzy matching fallback
4. Grouping (Gom màu theo style)
"""
# --- Process vector in query field ---
query_text = getattr(params, "query", None)
if query_text and query_vector is None:
query_vector = await create_embedding_async(query_text)
# --- Build filter clauses (OPTIMIZED ORDER) ---
where_clauses = _get_where_clauses(params)
where_sql = " AND ".join(where_clauses) if where_clauses else "1=1"
# --- Build SQL ---
if query_vector and len(query_vector) > 0:
v_str = "[" + ",".join(str(v) for v in query_vector) + "]"
sql = f"""
WITH top_sku_candidates AS (
SELECT
internal_ref_code,
product_name,
sale_price,
original_price,
master_color,
product_image_url,
product_image_url_thumbnail,
product_web_url,
description_text,
material,
material_group,
gender_by_product,
age_by_product,
season,
style,
fitting,
form_neckline,
form_sleeve,
product_line_vn,
product_color_name,
cosine_similarity(vector, {v_str}) as similarity_score
FROM shared_source.magento_product_dimension_with_text_embedding
WHERE {where_sql} AND vector IS NOT NULL
ORDER BY similarity_score DESC
LIMIT 50
)
SELECT
internal_ref_code,
ANY_VALUE(product_name) as product_name,
ANY_VALUE(sale_price) as sale_price,
ANY_VALUE(original_price) as original_price,
GROUP_CONCAT(DISTINCT master_color ORDER BY master_color SEPARATOR ', ') as available_colors,
ANY_VALUE(product_image_url) as product_image_url,
ANY_VALUE(product_image_url_thumbnail) as product_image_url_thumbnail,
ANY_VALUE(product_web_url) as product_web_url,
ANY_VALUE(description_text) as description_text,
ANY_VALUE(material) as material,
ANY_VALUE(material_group) as material_group,
ANY_VALUE(gender_by_product) as gender_by_product,
ANY_VALUE(age_by_product) as age_by_product,
ANY_VALUE(season) as season,
ANY_VALUE(style) as style,
ANY_VALUE(fitting) as fitting,
ANY_VALUE(form_neckline) as form_neckline,
ANY_VALUE(form_sleeve) as form_sleeve,
ANY_VALUE(product_line_vn) as product_line_vn,
MAX(similarity_score) as max_score
FROM top_sku_candidates
GROUP BY internal_ref_code
ORDER BY max_score DESC
LIMIT 10
""" # noqa: S608
else:
# ⚡ FALLBACK: FLEXIBLE KEYWORD SEARCH (OR + SCORING)
# Giải quyết case: User search "áo khoác nỉ" → DB có "Áo nỉ nam"
keywords = getattr(params, "keywords", None)
keyword_score_sql = ""
keyword_filter = ""
if keywords:
k_clean = _escape(keywords).lower().strip()
if k_clean:
words = k_clean.split()
# Build scoring expression: Each matched word = +1 point
# Example: "áo khoác nỉ" (3 words)
# - "Áo nỉ nam" matches 2/3 → Score = 2
# - "Áo khoác nỉ hoodie" matches 3/3 → Score = 3
score_terms = [
f"(CASE WHEN LOWER(product_name) LIKE '%{w}%' THEN 1 ELSE 0 END)"
for w in words
]
keyword_score_sql = f"({' + '.join(score_terms)}) as keyword_match_score"
# Minimum threshold: At least 50% of words must match
# Example: 3 words → need at least 2 matches (66%)
# 2 words → need at least 1 match (50%)
min_matches = max(1, len(words) // 2)
keyword_filter = f" AND ({' + '.join(score_terms)}) >= {min_matches}"
# Select clause with optional scoring
select_score = f", {keyword_score_sql}" if keyword_score_sql else ""
order_by = "keyword_match_score DESC, sale_price ASC" if keyword_score_sql else "sale_price ASC"
sql = f"""
SELECT
internal_ref_code,
ANY_VALUE(product_name) as product_name,
ANY_VALUE(sale_price) as sale_price,
ANY_VALUE(original_price) as original_price,
GROUP_CONCAT(DISTINCT master_color ORDER BY master_color SEPARATOR ', ') as available_colors,
ANY_VALUE(product_image_url) as product_image_url,
ANY_VALUE(product_image_url_thumbnail) as product_image_url_thumbnail,
ANY_VALUE(product_web_url) as product_web_url,
ANY_VALUE(description_text) as description_text,
ANY_VALUE(material) as material,
ANY_VALUE(material_group) as material_group,
ANY_VALUE(gender_by_product) as gender_by_product,
ANY_VALUE(age_by_product) as age_by_product,
ANY_VALUE(season) as season,
ANY_VALUE(style) as style,
ANY_VALUE(fitting) as fitting,
ANY_VALUE(form_neckline) as form_neckline,
ANY_VALUE(form_sleeve) as form_sleeve,
ANY_VALUE(product_line_vn) as product_line_vn
{select_score}
FROM shared_source.magento_product_dimension_with_text_embedding
WHERE {where_sql} {keyword_filter}
GROUP BY internal_ref_code
HAVING COUNT(*) > 0
ORDER BY {order_by}
LIMIT 10
""" # noqa: S608
# Log filter statistics
filter_info = f"Mode: {'Vector' if query_vector else 'Keyword'}, Filters: {len(where_clauses)}"
if where_clauses:
# Identify high-priority filters used
has_sku = any('internal_ref_code' in c or 'magento_ref_code' in c for c in where_clauses)
has_gender = any('gender_by_product' in c for c in where_clauses)
has_category = any('product_line_vn' in c for c in where_clauses)
priority_info = []
if has_sku:
priority_info.append("SKU")
if has_gender:
priority_info.append("Gender")
if has_category:
priority_info.append("Category")
if priority_info:
filter_info += f", Priority: {'+'.join(priority_info)}"
logger.info(f"📊 {filter_info}")
# Write SQL to file for debugging
try:
with open(r"d:\cnf\chatbot_canifa\backend\embedding.txt", "w", encoding="utf-8") as f:
f.write(sql)
except Exception as e:
logger.error(f"Failed to write SQL to embedding.txt: {e}")
return sql
\ No newline at end of file
backend/agent/tools/store_search_tool.py
View file @ 94484304
......@@ -20,28 +20,23 @@ class StoreSearchInput(BaseModel):
async def canifa_store_search(location: str) -> str:
"""
Tìm kiếm cửa hàng CANIFA theo địa điểm/khu vực.
Sử dụng khi khách hàng hỏi về cửa hàng tại một khu vực cụ thể.
"""
logger.info(f"🏪 [Store Search] Location: {location}")
logger.info(f"🏪 [Store Search] Location: {location} (StarRocks)")
try:
sr = get_db_connection()
# Clean location: bỏ prefix generic
clean = location.lower().strip()
for prefix in ["quận ", "huyện ", "tỉnh ", "thành phố ", "tp. ", "tp "]:
clean = clean.replace(prefix, "")
clean = clean.strip()
# Tách thành tokens, deduplicate (giữ thứ tự)
# VD: "hà đông, hà nội" → ["hà", "đông", "nội"]
tokens = list(dict.fromkeys(
t for t in clean.replace(',', ' ').split() if t.strip()
))
if not tokens:
return "Vui lòng cho em biết khu vực bạn muốn tìm cửa hàng CANIFA (ví dụ: Hoàng Mai, Cầu Giấy, Đà Nẵng...)."
return "Vui lòng cho em biết khu vực bạn muốn tìm cửa hàng CANIFA."
# Search trên concat tất cả cột địa chỉ
text_col = "LOWER(concat_ws(' ', store_name, address, city, state))"
def _build_sql(where_clause: str) -> str:
......@@ -54,39 +49,12 @@ async def canifa_store_search(location: str) -> str:
LIMIT 20
"""
# ═══════════════════════════════════════════════
# Step 1: AND tất cả tokens (strict match)
# "hà đông hà nội" → tokens ["hà","đông","nội"] → AND → 5 stores ✓
# ═══════════════════════════════════════════════
and_conds = [f"{text_col} LIKE '%{tk}%'" for tk in tokens]
results = await sr.execute_query_async(_build_sql(' AND '.join(and_conds)))
# ═══════════════════════════════════════════════
# Step 2: Fallback — Reverse LIKE
# Dùng chính DB làm từ điển địa danh:
# Kiểm tra tên quận/huyện/tỉnh nào trong DB XUẤT HIỆN trong input user
# "hà đông cầu giấy" chứa "hà đông" (city) + "cầu giấy" (city) → lấy CẢ 2
# ═══════════════════════════════════════════════
if not results and len(tokens) >= 2:
# Strip prefix khỏi city: "Quận Hà Đông" → "hà đông"
city_stripped = """LOWER(TRIM(REPLACE(REPLACE(REPLACE(REPLACE(REPLACE(
city, 'Quận ', ''), 'Huyện ', ''), 'Thành phố ', ''), 'Thị xã ', ''), 'TP. ', '')))"""
state_lower = "LOWER(TRIM(state))"
fallback_where = f"""
(LOCATE({city_stripped}, '{clean}') > 0 AND LENGTH({city_stripped}) > 1)
OR
(LOCATE({state_lower}, '{clean}') > 0 AND LENGTH({state_lower}) > 1)
"""
results = await sr.execute_query_async(_build_sql(fallback_where))
logger.info(f"📊 Store search: reverse-LIKE fallback for '{clean}'")
logger.info(f"📊 Store search: {len(results)} stores found for '{location}'")
if not results:
return f"Không tìm thấy cửa hàng CANIFA tại khu vực '{location}'. Khách hàng có thể liên hệ hotline 1800 6061 để được hỗ trợ tìm cửa hàng gần nhất."
return f"Không tìm thấy cửa hàng CANIFA tại khu vực '{location}'."
# Format kết quả rõ ràng cho LLM
lines = []
for r in results:
name = r.get("store_name", "")
......@@ -98,22 +66,11 @@ async def canifa_store_search(location: str) -> str:
t_open = r.get("time_open_today", "")
t_close = r.get("time_close_today", "")
store_info = f"🏪 {name}\n"
store_info += f" 📍 Địa chỉ: {addr}"
if city:
store_info += f", {city}"
if state:
store_info += f", {state}"
store_info += f"\n 📞 ĐT: {phone}"
store_info += f"\n 🕐 Lịch: {schedule}"
if t_open and t_close:
store_info += f" (Hôm nay: {t_open}-{t_close})"
store_info = f"🏪 {name}\n 📍 Địa chỉ: {addr}, {city}, {state}\n 📞 ĐT: {phone}\n 🕐 Lịch: {schedule} ({t_open}-{t_close})"
lines.append(store_info)
return f"Tìm thấy {len(results)} cửa hàng CANIFA tại khu vực '{location}':\n\n" + "\n\n".join(lines)
return f"Tìm thấy {len(results)} cửa hàng CANIFA tại '{location}':\n\n" + "\n\n".join(lines)
except Exception as e:
logger.error(f"❌ Error in canifa_store_search: {e}")
return "Tôi đang gặp khó khăn khi tìm kiếm cửa hàng. Bạn có thể liên hệ hotline 1800 6061 để được hỗ trợ."
return "Tôi đang gặp khó khăn khi tìm kiếm cửa hàng."
backend/agent/tools/tool_module/__init__.py 0 → 100644
View file @ 94484304
from .search_engine import SearchEngine
from .db_connector import DBConnector
from .stock_provider import enrich_with_stock
from .product_mapping import resolve_product_line, get_related_lines
from .pattern_detector import HardPatternDetector
from .size_message_builder import build_size_message
__all__ = [
"SearchEngine",
"DBConnector",
"enrich_with_stock",
"resolve_product_line",
"get_related_lines",
"HardPatternDetector",
"build_size_message"
]
backend/agent/tools/tool_module/db_connector.py 0 → 100644
View file @ 94484304
import logging
import re
import sqlite3
import aiosqlite
from typing import Optional
from common.starrocks_connection import get_db_connection
from common.sqlite_db import sqlite_db
from common.constants import SQLITE_DB_PATH
logger = logging.getLogger(__name__)
def translate_query(query: str) -> str:
"""
Dịch câu SQL đặc thù của StarRocks sang dạng mà SQLite hiểu được.
Mượn logic từ sqlite_mock.py của bro.
"""
# 1. Thay thế Placeholder (%s -> ?)
q = re.sub(r'(?<!%)%s(?!%)', '?', query)
# 2. Thay thế cấu trúc của StarRocks (table name)
# test_db.magento_... → sr__test_db__magento_...
q = re.sub(
r'(?:test_db\.)?`?magento_product_dimension_with_text_embedding`?',
r'sr__test_db__magento_product_dimension_with_text_embedding',
q
)
# 3. Vá các ngữ pháp (Dialect)
q = re.sub(r'ANY_VALUE\s*\(', r'MAX(', q, flags=re.IGNORECASE)
q = re.sub(r'\bIF\s*\(', r'IIF(', q, flags=re.IGNORECASE)
q = re.sub(r'NULLS\s+LAST', '', q, flags=re.IGNORECASE)
q = re.sub(r'\bILIKE\b', 'LIKE', q, flags=re.IGNORECASE)
q = re.sub(r'\bTRUE\b', '1', q, flags=re.IGNORECASE)
q = re.sub(r'\bFALSE\b', '0', q, flags=re.IGNORECASE)
# 4. FIND_IN_SET(%s, REPLACE(col, '|', ',')) > 0 → INSTR fallback
# SQLite không hỗ trợ FIND_IN_SET, thay bằng INSTR
q = re.sub(
r"FIND_IN_SET\s*\(\s*(\?)\s*,\s*REPLACE\s*\(\s*(\w+)\s*,\s*'[|]'\s*,\s*','\s*\)\s*\)\s*>\s*0",
r"('|' || \2 || '|') LIKE ('%|' || \1 || '|%')",
q, flags=re.IGNORECASE
)
logger.debug(f"📝 [SQLite Translated] {q[:200]}")
return q
class DBConnector:
"""
Bộ điều phối DB: StarRocks vs SQLite.
"""
def __init__(self, use_sqlite: bool = False):
self.use_sqlite = use_sqlite
async def execute_query(self, query: str, params: tuple = ()) -> list[dict]:
if self.use_sqlite:
t_query = translate_query(query)
logger.info(f"📁 [DBConnector] Using SQLite Mock | Query: {t_query[:100]}...")
try:
return await sqlite_db.fetch_all(t_query, params)
except Exception as e:
logger.error(f"❌ [DBConnector] SQLite Error: {e}")
return []
else:
logger.info(f"🚀 [DBConnector] Using StarRocks Production")
db = get_db_connection()
if not db:
logger.error("❌ [DBConnector] StarRocks Connection Failed!")
return []
try:
return await db.execute_query_async(query, params)
except Exception as e:
logger.error(f"❌ [DBConnector] StarRocks Error: {e}. Falling back to SQLite...")
# Tự động cứu bồ
t_query = translate_query(query)
return await sqlite_db.fetch_all(t_query, params)
backend/agent/tools/tool_module/pattern_detector.py 0 → 100644
View file @ 94484304
import re
from typing import Optional, Any
from dataclasses import dataclass
@dataclass
class PatternMatch:
pattern_type: str # "cheaper_than", "similar_to", "check_size", ...
reference_sku: Optional[str] = None
reference_product_line: Optional[str] = None
reference_price: Optional[float] = None
reference_color: Optional[str] = None
reference_size: Optional[str] = None
target_price_max: Optional[float] = None
target_price_min: Optional[float] = None
target_size: Optional[str] = None
target_color: Optional[str] = None
confidence: float = 0.0
raw_match: str = ""
class HardPatternDetector:
CHEAPER_PATTERNS = [r"(?:rẻ|giá)\s+(?:hơn|thấp hơn|xuống hơn)", r"có.*rẻ.*hơn"]
EXPENSIVE_PATTERNS = [r"(?:đắt|giá)\s+(?:hơn|cao hơn)", r"có.*đắt.*hơn"]
SIMILAR_PATTERNS = [r"(?:giống|tương tự|như|cùng)\s+(?:như|với)?\s+(?:cái|chiếc)?\s*(?:đó|trước|ban\s*nãy|cạnh|là)?"]
SIZE_CHECK_PATTERNS = [r"(?:còn|có)\s+size\s+([A-Z0-9]+)"]
COLOR_CHECK_PATTERNS = [r"(?:màu\s+)?(\w+)\s+(?:còn|có)"]
STOCK_CHECK_PATTERNS = [r"(?:còn|có)\s+(?:hàng|kho)\s+(?:không|ko|kh)"]
SKU_PATTERN = r"\b([A-Z0-9]{4,}-[A-Z0-9]{3,}(?:-[A-Z0-9]{3,})?)\b"
def detect(self, query: str, context_products: list[dict] = None) -> PatternMatch:
query_lower = query.lower().strip()
match = PatternMatch(pattern_type="unknown")
if any(re.search(p, query_lower) for p in self.CHEAPER_PATTERNS):
match.pattern_type = "cheaper_than"
match.confidence = 0.9
elif any(re.search(p, query_lower) for p in self.EXPENSIVE_PATTERNS):
match.pattern_type = "more_expensive_than"
match.confidence = 0.9
elif any(re.search(p, query_lower) for p in self.SIMILAR_PATTERNS):
match.pattern_type = "similar_to"
match.confidence = 0.8
elif sm := re.search(self.SIZE_CHECK_PATTERNS[0], query_lower):
match.pattern_type = "check_size"
match.target_size = sm.group(1).upper()
# Populate reference info if available
if context_products and match.pattern_type != "unknown":
last_p = context_products[-1]
match.reference_sku = last_p.get("code") or last_p.get("magento_ref_code")
match.reference_price = float(last_p.get("price") or 0)
match.reference_product_line = last_p.get("product_line_vn")
if match.pattern_type == "cheaper_than":
match.target_price_max = match.reference_price * 0.95
elif match.pattern_type == "more_expensive_than":
match.target_price_min = match.reference_price * 1.05
return match
backend/agent/tools/tool_module/product_mapping.py 0 → 100644
View file @ 94484304
"""
Product Line Mapping - Đồng bộ từ Lead Stage
"""
PRODUCT_LINE_MAP: dict[str, list[str]] = {
"Áo Sơ mi": ["áo sơ mi", "áo công sở", "áo đi làm", "sơ mi", "sơmi", "áo sơmi"],
"Áo Polo": ["áo polo", "áo cổ bẻ", "polo"],
"Áo phông": ["áo phông", "áo thun", "áo thun ngắn tay", "áo cổ v", "áo cổ tym", "áo cộc tay"],
"Áo nỉ có mũ": ["áo nỉ có mũ", "áo hoodie", "hoodie"],
"Áo nỉ": ["áo nỉ", "áo sweater", "sweater"],
"Áo mặc nhà": ["áo mặc nhà", "áo ngủ", "áo ở nhà"],
"Áo lót": ["áo lót", "áo ngực", "áo quây", "áo lót nữ", "áo lót nam", "áo lót trẻ em"],
"Áo len gilet": ["áo len gilet", "áo gile len"],
"Áo len": ["áo len", "áo len dài tay"],
"Áo kiểu": ["áo kiểu", "áo điệu", "áo nữ tính"],
"Áo khoác sợi": ["áo khoác sợi"],
"Áo khoác nỉ không mũ": ["áo khoác nỉ không mũ", "áo khoác sweater"],
"Áo khoác nỉ có mũ": ["áo khoác nỉ có mũ", "áo khoác hoodie", "áo khoác nỉ"],
"Áo khoác lông vũ": ["áo khoác lông vũ", "áo phao lông vũ", "áo lông vũ"],
"Áo khoác gió": ["áo khoác gió", "áo gió", "áo khoác mỏng"],
"Áo khoác gilet chần bông": ["áo khoác gilet chần bông", "áo khoác gilet trần bông", "áo gilet chần bông", "áo gilet trần bông"],
"Áo khoác gilet": ["áo khoác gilet", "áo gile", "gile"],
"Áo khoác dạ": ["áo khoác dạ", "áo dạ"],
"Áo khoác dáng ngắn": ["áo khoác dáng ngắn", "áo khoác croptop"],
"Áo khoác chống nắng": ["áo khoác chống nắng", "áo chống nắng"],
"Áo khoác chần bông": ["áo khoác chần bông", "áo khoác trần bông", "áo chần bông", "áo trần bông", "áo phao"],
"Áo khoác": ["áo khoác", "áo ấm", "áo rét"],
"Áo giữ nhiệt": ["áo giữ nhiệt", "áo tản nhiệt", "áo heattech"],
"Áo bra active": ["áo bra active", "áo bra", "bra", "áo tập", "áo thể thao"],
"Áo Body": ["áo body", "áo croptop", "croptop", "baby tee", "áo lửng", "áo dáng ngắn", "áo ôm"],
"Áo ba lỗ": ["áo ba lỗ", "áo sát nách", "tanktop", "tank top", "áo dây", "áo 2 dây", "áo hai dây"],
"Váy liền": ["váy liền", "đầm", "váy công sở", "đầm công sở", "váy liền thân", "đầm suông", "váy dài", "váy body"],
"Chân váy": ["chân váy", "váy maxi", "váy midi", "chân váy dài", "chân váy chữ a", "chân váy công sở", "váy ngắn"],
"Quần giả váy": ["quần giả váy", "quần váy", "skort"],
"Quần soóc": ["quần soóc", "quần đùi", "quần short", "quần lửng", "quần ngố", "short", "quần đùi nam", "quần đùi nữ"],
"Quần nỉ": ["quần nỉ", "quần jogger", "quần ống bo chun", "jogger", "quần thể thao"],
"Quần mặc nhà": ["quần mặc nhà", "quần ngủ", "quần đùi mặc nhà"],
"Quần lót đùi": ["quần lót đùi", "quần sịp đùi", "quần boxer", "boxer", "sịp đùi", "quần xì đùi"],
"Quần lót tam giác": ["quần lót tam giác", "quần sịp tam giác", "quần brief", "brief", "sịp tam giác", "quần xì tam giác"],
"Quần lót": ["quần lót", "quần chip", "quần sịp", "quần trong", "quần nhỏ", "quần xơ lít", "quần xì", "sịp", "chip", "đồ lót"],
"Quần leggings mặc nhà": ["quần leggings mặc nhà", "quần legging mặc nhà"],
"Quần leggings": ["quần leggings", "leggings", "quần legging", "legging", "quần thun ôm"],
"Quần Khaki": ["quần khaki", "quần âu", "quần vải", "quần tây", "quần công sở", "quần đi làm", "quần âu nam", "quần âu nữ", "quần kaki"],
"Quần jean": ["quần jean", "quần bò", "quần jeans", "denim", "jeans", "bò", "jean", "quần dzin"],
"Quần giữ nhiệt": ["quần giữ nhiệt", "quần heattech"],
"Quần dài": ["quần dài", "quần suông", "quần ống rộng", "quần ống suông", "quần lưng thun"],
"Quần culottes": ["quần culottes", "culottes", "quần lửng ống rộng"],
"Quần Body": ["quần body", "quần ôm"],
"Pyjama": ["pyjama", "pajama", "đồ pijama"],
"Mũ": ["mũ", "nón", "phụ kiện Canifa"],
"Tất": ["tất", "vớ", "bao chân", "vớ chân", "tất chân"],
"Quần tất": ["quần tất", "quần vớ", "tất quần"],
"Mũ thể thao": ["mũ thể thao", "mũ snapback", "mũ lưỡi trai", "cap"],
"Khẩu trang": ["khẩu trang", "mask", "mặt nạ vải"],
"Túi xách": ["túi xách", "túi"],
}
SYNONYM_TO_DB: dict[str, str] = {}
for db_value, synonyms in PRODUCT_LINE_MAP.items():
for syn in synonyms:
SYNONYM_TO_DB[syn.lower()] = db_value
RELATED_LINES: dict[str, list[str]] = {
"Áo bra active": ["Áo lót"],
"Áo lót": ["Áo bra active"],
"Quần lót": ["Quần lót đùi", "Quần lót tam giác"],
"Quần lót đùi": ["Quần lót", "Quần lót tam giác"],
"Quần lót tam giác": ["Quần lót", "Quần lót đùi"],
}
def get_related_lines(product_line: str) -> list[str]:
return [product_line] + RELATED_LINES.get(product_line, [])
def resolve_product_line(raw_value: str) -> list[str]:
parts = [p.strip() for p in raw_value.split("/") if p.strip()]
resolved = []
for part in parts:
mapped = SYNONYM_TO_DB.get(part.lower())
if mapped:
resolved.append(mapped)
else:
resolved.append(part)
return resolved
backend/agent/tools/tool_module/search_engine.py 0 → 100644
View file @ 94484304
import json
import logging
import os as _os
import re
import sqlite3
import time
import httpx
import asyncio
from typing import List, Tuple, Dict, Any, Optional
from pydantic import BaseModel, Field
# Core module imports từ local tool_module
from .db_connector import DBConnector
from .stock_provider import enrich_with_stock
from .product_mapping import get_related_lines, resolve_product_line
from .pattern_detector import HardPatternDetector
from .size_message_builder import build_size_message
# Constants
from common.constants import SQLITE_DB_PATH
logger = logging.getLogger(__name__)
# Global instances
_hard_pattern_detector = HardPatternDetector()
TABLE_NAME = "test_db.magento_product_dimension_with_text_embedding"
SELECT_COLUMNS = """
internal_ref_code,
magento_ref_code,
product_color_code,
product_name,
master_color,
product_color_name,
product_image_url_thumbnail,
product_web_url,
sale_price,
original_price,
COALESCE(discount_amount, 0) AS discount_amount,
ROUND(CASE WHEN original_price > 0
THEN ((original_price - sale_price) / original_price * 100)
ELSE 0 END, 0) AS discount_percent,
age_by_product,
gender_by_product,
product_line_vn,
COALESCE(quantity_sold, 0) AS quantity_sold,
COALESCE(is_new_product, 0) AS is_new_product,
size_scale,
description_text
"""
# ==============================================================================
# AGENTIC DUAL-LANE SEARCH: LOGIC & EVALUATION (ƯU/NHƯỢC ĐIỂM)
# ==============================================================================
"""
Cơ chế Dual-Lane Search (Literal + Inferred) giúp tối ưu hóa khả năng tìm kiếm:
- Ưu điểm:
1. Inferred Lane: Đảm bảo độ chính xác tuyệt đối với các bộ lọc thuộc tính (màu, giá, giới tính).
2. Literal Lane: Bắt được các từ khóa cảm xúc, phong cách (đi tiệc, đi chơi) mà DB không có cột riêng.
3. Fallback Cascade: Giảm thiểu tối đa tình trạng "không tìm thấy kết quả" bằng cách nới lỏng tiêu chí.
- Nhược điểm:
1. Keyword Noise: Có thể bắt nhầm từ khóa trong mô tả sản phẩm (ví dụ: "không dành cho...")
2. Latency: Cần tối ưu hóa query StarRocks khi thực hiện nhiều tầng fallback.
"""
# ==============================================================================
# LOGIC BỔ TRỢ (DI CƯ TỪ LEAD STAGE)
# ==============================================================================
TAGS_TO_OCCASION: dict[str, str] = {
"occ:di_lam": "di_lam",
"occ:di_choi": "di_choi",
"occ:di_tiec": "di_tiec",
"occ:di_hoc": "di_choi",
"occ:mac_nha": "mac_nha",
"occ:the_thao": "the_thao",
"occ:di_bien": "di_choi",
"occ:du_lich": "di_choi",
"occ:da_ngoai": "di_choi",
"occ:di_ngu": "mac_nha",
"occ:hang_ngay": "hang_ngay",
}
EVENT_FALLBACKS: list[dict] = [
{
"patterns": ["30/4", "30-4", "giai phong", "giải phóng"],
"master_color": "Màu đỏ",
"tags": ["Màu đỏ", "Đi chơi / dạo phố"],
"keywords": ["cờ đỏ sao vàng", "30/4"],
},
{
"patterns": ["2/9", "2-9", "quoc khanh", "quốc khánh"],
"master_color": "Màu đỏ",
"tags": ["Màu đỏ", "Đi chơi / dạo phố"],
"keywords": ["quốc khánh", "lễ quốc khánh"],
},
]
def _normalize_text(text: str) -> str:
import unicodedata
lowered = text.lower().strip()
return "".join(ch for ch in unicodedata.normalize("NFD", lowered) if unicodedata.category(ch) != "Mn")
def _apply_event_fallback(inf: Any, text: str) -> bool:
if not text: return False
norm = _normalize_text(text)
for rule in EVENT_FALLBACKS:
if any(pat in norm for pat in rule["patterns"]):
if not inf.master_color: inf.master_color = rule["master_color"]
if not inf.tags: inf.tags = list(rule["tags"])
if not inf.keywords: inf.keywords = list(rule["keywords"])
return True
return False
# ==============================================================================
# SEARCH MODELS
# ==============================================================================
class LiteralSearch(BaseModel):
raw_text: str = ""
class InferredSearch(BaseModel):
product_line_vn: List[str] = Field(default_factory=list)
gender_by_product: Optional[str] = None
age_by_product: Optional[str] = None
master_color: Optional[str] = None
tags: List[str] = Field(default_factory=list)
keywords: List[str] = Field(default_factory=list)
price_min: Optional[int] = None
price_max: Optional[int] = None
discount_min: Optional[int] = None
discount_max: Optional[int] = None
discovery_mode: Optional[str] = None
size: Optional[str] = None
# ==============================================================================
# SQL BUILDER HELPERS
# ==============================================================================
def _build_fixed_clauses(inf: InferredSearch, params: list) -> list[str]:
clauses = []
if inf.product_line_vn:
lines = []
for line in inf.product_line_vn:
if not line: continue
resolved = resolve_product_line(line)
for r in resolved: lines.extend(get_related_lines(r))
if lines:
placeholders = ", ".join(["%s"] * len(lines))
params.extend(lines)
clauses.append(f"product_line_vn IN ({placeholders})")
if inf.gender_by_product:
g = inf.gender_by_product.lower()
# DB values: men, women, boy, girl, unisex, others
if g in ["women", "nu", "female", "nữ", "woman"]:
genders = ["women", "unisex"]
elif g in ["men", "nam", "male", "man"]:
genders = ["men", "unisex"]
elif g in ["boy", "bé trai", "be_trai"]:
genders = ["boy", "unisex"]
elif g in ["girl", "bé gái", "be_gai"]:
genders = ["girl", "unisex"]
else:
genders = [g, "unisex"]
params.extend(genders)
clauses.append(f"gender_by_product IN ({', '.join(['%s']*len(genders))})")
if inf.age_by_product:
# DB values: adult, kid, others
age = inf.age_by_product.lower()
if age in ["children", "child", "kid", "kids", "trẻ em", "tre_em", "bé"]:
age = "kid"
elif age in ["adult", "adults", "người lớn"]:
age = "adult"
params.append(age)
clauses.append("age_by_product = %s")
if inf.master_color:
params.append(f"%{inf.master_color}%"); params.append(f"%{inf.master_color}%")
clauses.append("(master_color LIKE %s OR product_color_name LIKE %s)")
if inf.price_min is not None:
params.append(inf.price_min); clauses.append("sale_price >= %s")
if inf.price_max is not None:
params.append(inf.price_max); clauses.append("sale_price <= %s")
if inf.size:
params.append(inf.size.upper())
clauses.append("FIND_IN_SET(%s, REPLACE(size_scale, '|', ',')) > 0")
return clauses
def _build_exclusion_clauses(keywords: list, params: list) -> list[str]:
clauses = []
kws = " ".join(keywords).lower()
if any(x in kws for x in ["đông", "lạnh", "winter"]):
clauses.append("product_line_vn NOT IN (%s, %s)")
params.extend(["Quần soóc", "Áo ba lỗ"])
return clauses
def _build_full_query(fixed: list, search: str = None, exclusions: list = None) -> str:
all_c = list(fixed)
if search: all_c.append(search)
if exclusions: all_c.extend(exclusions)
where = " AND ".join(all_c) if all_c else "1=1"
return f"SELECT {SELECT_COLUMNS} FROM {TABLE_NAME} WHERE {where} ORDER BY quantity_sold DESC LIMIT 150"
# ==============================================================================
# MAIN SEARCH ENGINE
# ==============================================================================
class SearchEngine:
def __init__(self, use_sqlite: bool = False):
self.db = DBConnector(use_sqlite=use_sqlite)
async def _dual_lane_search(self, literal: str, inf: InferredSearch) -> Tuple[List[Dict], int, str]:
# Lane 1: Literal LIKE Search
literal_products = []
if literal:
stop_words = {"tìm", "mua", "có", "không", "nào", "cho", "xin", "nhé", "combo", "bộ", "set"}
words = [w.lower() for w in literal.split() if w.lower() not in stop_words]
if words:
params_lit = []
clauses_lit = []
for w in words:
params_lit.extend([f"%{w}%", f"%{w}%"])
clauses_lit.append("(LOWER(product_name) LIKE %s OR LOWER(description_text) LIKE %s)")
# Áp dụng cứng Gender/Age
where_lit = " AND ".join(clauses_lit)
if inf.gender_by_product:
params_lit.append(inf.gender_by_product); where_lit += " AND gender_by_product = %s"
if inf.age_by_product:
params_lit.append(inf.age_by_product); where_lit += " AND age_by_product = %s"
sql_lit = f"SELECT {SELECT_COLUMNS} FROM {TABLE_NAME} WHERE {where_lit} ORDER BY quantity_sold DESC LIMIT 50"
literal_products = await self.db.execute_query(sql_lit, tuple(params_lit))
# Lane 2: Inferred Structured Search
params_inf = []
fixed_inf = _build_fixed_clauses(inf, params_inf)
ex_params = []
exclusions = _build_exclusion_clauses(inf.keywords, ex_params)
sql_inf = _build_full_query(fixed_inf, None, exclusions)
inferred_products = await self.db.execute_query(sql_inf, tuple(params_inf + ex_params))
# Merge & Dedup (Ưu tiên Inferred trước để chuẩn phân loại)
seen = set()
merged = []
for p in inferred_products + literal_products:
code = p.get("internal_ref_code")
if code and code not in seen:
seen.add(code); merged.append(p)
if merged: return merged[:10], 1, "Success (Dual-Lane)"
return await self._fallback_cascade(inf)
async def _fallback_cascade(self, inf: InferredSearch) -> Tuple[List[Dict], int, str]:
ex_params = []
exclusions = _build_exclusion_clauses(inf.keywords, ex_params)
# Tier 4: Drop Gender
if inf.gender_by_product:
saved = inf.gender_by_product; inf.gender_by_product = None
params = []
fixed = _build_fixed_clauses(inf, params)
products = await self.db.execute_query(_build_full_query(fixed, None, exclusions), tuple(params + ex_params))
inf.gender_by_product = saved
if products: return products[:10], 4, "Fallback (Dropped Gender)"
# Tier 5-6: Price Relaxation
if inf.price_max:
orig_max = inf.price_max
for mult, tier in [(1.5, 5), (2.0, 6)]:
inf.price_max = int(orig_max * mult)
params = []
fixed = _build_fixed_clauses(inf, params)
products = await self.db.execute_query(_build_full_query(fixed, None, exclusions), tuple(params + ex_params))
if products:
inf.price_max = orig_max
return products[:10], tier, f"Fallback (Price relaxed {mult}x)"
inf.price_max = orig_max
# Tier 7: Drop Price entirely
saved_max, saved_min = inf.price_max, inf.price_min
inf.price_max = inf.price_min = None
params = []
fixed = _build_fixed_clauses(inf, params)
products = await self.db.execute_query(_build_full_query(fixed, None, exclusions), tuple(params + ex_params))
inf.price_max, inf.price_min = saved_max, saved_min
if products: return products[:10], 7, "Fallback (Unlimited Price)"
return [], 0, "No results found"
async def _enrich_with_outfit(self, products: list) -> list:
if not products or not _os.path.exists(SQLITE_DB_PATH): return products
top_codes = [p.get("internal_ref_code") for p in products[:3] if p.get("internal_ref_code")]
if not top_codes: return products
try:
conn = sqlite3.connect(SQLITE_DB_PATH)
conn.row_factory = sqlite3.Row
cursor = conn.cursor()
placeholders = ",".join(["?"] * len(top_codes))
outfits = cursor.execute(f"SELECT * FROM pg__dashboard_canifa__ai_outfit_product_matches WHERE anchor_product_code IN ({placeholders})", top_codes).fetchall()
conn.close()
outfit_map = {}
for row in outfits:
anchor = row["anchor_product_code"]
outfit_map.setdefault(anchor, []).append({
"code": row["match_product_code"],
"name": row["match_product_name"],
"role": row["match_role"],
"reason": row["ai_reason"]
})
for p in products:
p["outfit_recommendations"] = outfit_map.get(p.get("internal_ref_code"), [])
except Exception as e: logger.error(f"Outfit error: {e}")
return products
async def search(self, literal: str, inferred: Dict[str, Any], check_stock: bool = True) -> Dict[str, Any]:
inf = InferredSearch(**inferred)
_apply_event_fallback(inf, literal)
# Hard Pattern
pattern = _hard_pattern_detector.detect(literal)
if pattern.pattern_type == "cheaper_than" and pattern.target_price_max:
inf.price_max = int(pattern.target_price_max)
# Underwear Override
is_underwear = any(line in ["Quần lót", "Áo lót", "Sịp", "Bra"] for line in inf.product_line_vn)
if is_underwear: inf.gender_by_product = inf.age_by_product = None
products, tier, message = await self._dual_lane_search(literal, inf)
if products:
if check_stock: products = await enrich_with_stock(products)
products = await self._enrich_with_outfit(products)
for p in products:
raw_size = p.get("size_scale", "")
parsed = [s.strip() for s in str(raw_size).replace("[", "").replace("]", "").replace('"', '').split(",") if s.strip()]
p["size_message"] = build_size_message(p.get("gender_by_product", ""), p.get("product_line_vn", ""), parsed, p.get("description_text", ""))
return {
"status": "success" if products else "no_results",
"tier": tier, "message": message, "count": len(products), "results": products
}
backend/agent/tools/tool_module/size_message_builder.py 0 → 100644
View file @ 94484304
import logging
logger = logging.getLogger(__name__)
SIZE_MAPPING = {
"NU": {
"XS": "Cao 1m47-1m53, Nặng 38-43kg",
"S": "Cao 1m50-1m55, Nặng 41-46kg",
"M": "Cao 1m55-1m63, Nặng 47-52kg",
"L": "Cao 1m60-1m65, Nặng 53-58kg",
"XL": "Cao 1m62-1m66, Nặng 59-64kg",
},
"NAM": {
"S": "Cao 1m62-1m68, Nặng 57-62kg",
"M": "Cao 1m69-1m73, Nặng 63-67kg",
"L": "Cao 1m71-1m75, Nặng 68-72kg",
"XL": "Cao 1m73-1m77, Nặng 73-77kg",
"XXL": "Cao 1m75-1m79, Nặng 78-82kg",
},
"QUAN_NU": {
"26": "Vòng eo 65cm, Vòng mông 79-87cm",
"27": "Vòng eo 67.5cm, Vòng mông 81-89cm",
"28": "Vòng eo 70cm, Vòng mông 84-92cm",
"29": "Vòng eo 72.5cm, Vòng mông 86-94cm",
"30": "Vòng eo 75cm, Vòng mông 89-97cm",
},
"QUAN_NAM": {
"29": "Vòng eo 79.5cm, Vòng mông 96.5cm",
"30": "Vòng eo 82cm, Vòng mông 99cm",
"31": "Vòng eo 84.5cm, Vòng mông 101.5cm",
"32": "Vòng eo 87cm, Vòng mông 104cm",
"33": "Vòng eo 89cm, Vòng mông 106.5cm",
},
"TRE_EM": {
"90": "Dành cho bé 2Y, Cao 90cm, 10-13kg",
"100": "Dành cho bé 3-4Y, Cao 100cm, 14-17kg",
"110": "Dành cho bé 4-5Y, Cao 107-113cm, 18-23kg",
"120": "Dành cho bé 6-7Y, Cao 120cm, 24-29kg",
"130": "Dành cho bé 8Y, Cao 130cm, 29-33kg",
"140": "Dành cho bé 9-11Y, Cao 137-145cm, 33-39kg",
"150": "Dành cho bé 11-12Y, Cao 150cm, 39-45kg",
"160": "Dành cho bé 13-14Y, Cao 160cm, 45-52kg",
},
"UNISEX": {
"XXS": "Cao 1m55-1m63, Nặng 47-52kg",
"XS": "Cao 1m60-1m65, Nặng 53-58kg",
"S": "Cao 1m62-1m68, Nặng 57-62kg",
"M": "Cao 1m69-1m73, Nặng 63-67kg",
"L": "Cao 1m71-1m75, Nặng 68-72kg",
"XL": "Cao 1m73-1m77, Nặng 73-77kg",
"XXL": "Cao 1m75-1m79, Nặng 79-82kg",
}
}
def determine_table_key(gender: str, product_line: str) -> str:
gender = (gender or "").lower().strip()
product_line = (product_line or "").lower().strip()
is_jeans_or_khaki = any(x in product_line for x in ["jean", "khaki", "kaki", "quần âu", "quần tây"])
is_bottom = "quần" in product_line
if gender in ["boy", "girl", "kid", "bé trai", "bé gái", "trẻ em"]: return "TRE_EM"
if gender == "unisex": return "UNISEX"
if is_bottom and is_jeans_or_khaki:
if gender in ["women", "nu", "nữ", "female"]: return "QUAN_NU"
elif gender in ["men", "nam", "male"]: return "QUAN_NAM"
if gender in ["women", "nu", "nữ", "female"]: return "NU"
elif gender in ["men", "nam", "male"]: return "NAM"
return ""
def build_size_message(gender: str, product_line: str, sizes: list[str], description: str = "") -> str:
if not sizes: return ""
table_key = determine_table_key(gender, product_line)
if not table_key: return ""
mapping = SIZE_MAPPING.get(table_key, {})
size_lines = [f"Size {s.strip().upper()} ({mapping[s.strip().upper()]})" for s in sizes if str(s).strip().upper() in mapping]
if not size_lines: return ""
obj_str = {"NAM":"Sản phẩm Nam", "NU":"Sản phẩm Nữ", "TRE_EM":"Sản phẩm Trẻ em", "UNISEX":"Sản phẩm Unisex", "QUAN_NU":"Quần Nữ", "QUAN_NAM":"Quần Nam"}.get(table_key, "Sản phẩm")
base_msg = f"{obj_str}. Các size ĐANG CÒN HÀNG: {'; '.join(size_lines)}."
desc_lower = (description or "").lower()
if any(x in desc_lower for x in ["dáng ôm", "slim fit", "body"]):
base_msg += "\n\n🔥 Mẹo: Form ôm, nếu thích thoải mái nên nhích lên 1 size."
elif any(x in desc_lower for x in ["oversize", "rộng rãi"]):
base_msg += "\n\n🔥 Mẹo: Form rộng, nên lấy đúng size để lên dáng thụng đẹp."
return base_msg
backend/agent/tools/tool_module/stock_provider.py 0 → 100644
View file @ 94484304
import time
import httpx
import logging
logger = logging.getLogger(__name__)
CANIFA_STOCK_API = "https://canifa.com/v1/middleware/stock_get_stock_list_parent"
_STOCK_TIMEOUT = 2.0 # seconds
async def fetch_stock_batch(base_codes: list[str]) -> dict[str, list[dict]]:
"""
Gọi API Canifa để lấy thông tin tồn kho cho một danh sách SKU.
"""
if not base_codes:
return {}
sku_string = ",".join(base_codes)
url = f"{CANIFA_STOCK_API}?skus={sku_string}"
t0 = time.time()
try:
async with httpx.AsyncClient(timeout=_STOCK_TIMEOUT) as client:
resp = await client.get(url)
resp.raise_for_status()
data = resp.json()
elapsed = round((time.time() - t0) * 1000)
stock_map: dict[str, list[dict]] = {}
results = data.get("result", [])
if isinstance(results, list):
for item in results:
sku_full = item.get("sku", "")
qty = item.get("qty", 0) or 0
if not sku_full:
continue
# SKU thường có định dạng BASE-COLOR-SIZE
parts = sku_full.rsplit("-", 1)
if len(parts) == 2:
color_code = parts[0]
size = parts[1]
status = "còn hàng" if qty > 0 else "hết hàng"
stock_map.setdefault(color_code, []).append({
"size": size, "qty": qty, "status": status
})
logger.info(f"📦 [StockAPI] {len(base_codes)} SKUs -> {len(stock_map)} colors in {elapsed}ms")
return stock_map
except Exception as e:
logger.warning(f"⚠️ [StockAPI] Error or Timeout: {e}")
return {}
async def enrich_with_stock(products: list[dict]) -> list[dict]:
"""
Lọc danh sách sản phẩm, chỉ giữ lại những món còn hàng.
"""
if not products:
return []
# Lấy danh sách base code duy nhất (ví dụ: 8AT23S001)
base_codes = list({p.get("internal_ref_code", "") for p in products if p.get("internal_ref_code")})
if not base_codes:
return products
stock_map = await fetch_stock_batch(base_codes)
if not stock_map:
return products # Nếu API lỗi thì cứ trả về hết để AI tư vấn
in_stock = []
for p in products:
# product_color_code là mã màu cụ thể (ví dụ: 8AT23S001-SB001)
color_code = p.get("product_color_code", "")
if color_code in stock_map:
sizes_detail = stock_map[color_code]
total_qty = sum(s["qty"] for s in sizes_detail)
if total_qty > 0:
p["_stock_detail"] = sizes_detail
p["_total_qty"] = total_qty
in_stock.append(p)
# Nếu lọc xong mà trắng tay thì trả về 5 thằng đầu tiên để AI còn có cái mà nói
return in_stock if in_stock else products[:5]
backend/common/constants.py 0 → 100644
View file @ 94484304
"""
constants.py — Single Source of Truth cho toàn bộ hằng số dùng chung.
QUY TẮC:
- Hằng số viết HOA toàn bộ (UPPERCASE).
- Đặt ngay dưới phần import.
- Mọi file khác import từ đây, KHÔNG hardcode lại.
"""
import os
# ═══════════════════════════════════════════════════════════════
# DATABASE PATHS
# ═══════════════════════════════════════════════════════════════
_BACKEND_DIR = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
_PROJECT_ROOT = os.path.dirname(_BACKEND_DIR) # D:\cnf\chatbot_canifa
SQLITE_DB_PATH: str = os.path.join(_PROJECT_ROOT, "preference", "common", "database", "canifa_ai_dump.sqlite")
# ═══════════════════════════════════════════════════════════════
# SQLITE TABLE NAMES (mapping schema.table → SQLite table name)
# ═══════════════════════════════════════════════════════════════
TABLE_OUTFIT_RULES = "pg__dashboard_canifa__chatbot_fashion_rules"
TABLE_FASHION_RULES = "pg__dashboard_canifa__chatbot_fashion_rules"
TABLE_OUTFIT_MATCHES = "pg__dashboard_canifa__ai_outfit_product_matches"
TABLE_PRODUCTS = "sr__test_db__magento_product_dimension_with_text_embedding"
# ═══════════════════════════════════════════════════════════════
# GENDER MAPPING (Vietnamese / English → DB key)
# ═══════════════════════════════════════════════════════════════
SQLITE_GENDER_MAP: dict[str, str] = {
"nu": "women",
"nam": "men",
"be_gai": "girl",
"be_trai": "boy",
"unisex": "unisex",
}
# ═══════════════════════════════════════════════════════════════
# KID DETECTION KEYWORDS
# ═══════════════════════════════════════════════════════════════
KID_SAFE_KEYWORDS: list[str] = [
"kid", "trẻ em", "bé trai", "bé gái", "be_gai", "be_trai", "tre_em",
]
def normalize_gender(gender: str) -> str:
"""Chuẩn hoá chuỗi gender về key nội bộ (nu/nam/be_gai/be_trai/unisex)."""
g = (gender or "").lower().strip()
if g in ("nu", "nữ", "women", "female", "woman"):
return "nu"
if g in ("nam", "men", "male", "man"):
return "nam"
if any(k in g for k in ["bé gái", "be_gai", "girl", "be gai"]):
return "be_gai"
if any(k in g for k in ["bé trai", "be_trai", "boy", "be trai"]):
return "be_trai"
return "unisex"
backend/common/conversation_manager.py
View file @ 94484304
......@@ -41,7 +41,7 @@ class ConversationManager:
"""Create the chat history table if it doesn't exist"""
try:
pool = await self._get_pool()
async with pool.connection() as conn:
async with pool.connection(timeout=2.0) as conn:
async with conn.cursor() as cursor:
# Set timezone to Vietnam for this session
await cursor.execute("SET timezone = 'Asia/Ho_Chi_Minh'")
......@@ -81,7 +81,7 @@ class ConversationManager:
vietnam_tz = timezone(timedelta(hours=7))
timestamp = datetime.now(vietnam_tz)
# Transaction block: atomic insert
async with pool.connection() as conn:
async with pool.connection(timeout=2.0) as conn:
async with conn.cursor() as cursor:
# Set timezone to Vietnam for this session
await cursor.execute("SET timezone = 'Asia/Ho_Chi_Minh'")
......@@ -175,7 +175,7 @@ class ConversationManager:
final_query = sql.SQL(" ").join(query_parts)
pool = await self._get_pool()
async with pool.connection() as conn, conn.cursor() as cursor:
async with pool.connection(timeout=2.0) as conn, conn.cursor() as cursor:
# Set timezone to Vietnam for this session
await cursor.execute("SET timezone = 'Asia/Ho_Chi_Minh'")
......@@ -232,7 +232,7 @@ class ConversationManager:
end_of_day = now.replace(hour=23, minute=59, second=59, microsecond=999999)
pool = await self._get_pool()
async with pool.connection() as conn:
async with pool.connection(timeout=2.0) as conn:
async with conn.cursor() as cursor:
query = sql.SQL("""
UPDATE {table}
......@@ -255,7 +255,7 @@ class ConversationManager:
"""Clear all chat history for an identity"""
try:
pool = await self._get_pool()
async with pool.connection() as conn:
async with pool.connection(timeout=2.0) as conn:
async with conn.cursor() as cursor:
query = sql.SQL("DELETE FROM {table} WHERE identity_key = %s").format(
table=sql.Identifier(self.table_name)
......
backend/common/sqlite_db.py 0 → 100644
View file @ 94484304
import logging
import os
from contextlib import asynccontextmanager
from typing import Optional
import aiosqlite
logger = logging.getLogger(__name__)
class SQLiteDBManager:
"""
Quản lý kết nối tới file SQLite database (canifa_ai_dump.sqlite).
Được thiết kế theo Singleton pattern để dùng thống nhất trong toàn bộ App FastAPI.
Đã được chuyển đổi sang aiosqlite (bất đồng bộ).
"""
_instance = None
_db_path = None
def __new__(cls):
if cls._instance is None:
cls._instance = super(SQLiteDBManager, cls).__new__(cls)
from common.constants import SQLITE_DB_PATH
cls._instance._db_path = SQLITE_DB_PATH
return cls._instance
@property
def db_path(self) -> str:
return self._db_path
@db_path.setter
def db_path(self, path: str):
self._db_path = path
async def get_connection(self) -> aiosqlite.Connection:
"""
Khởi tạo và trả về 1 kết nối aiosqlite.
"""
if not os.path.exists(self._db_path):
logger.warning(f"[SQLite] Cảnh báo: File database không tồn tại tại {self._db_path}")
conn = await aiosqlite.connect(self._db_path)
# Row factory giúp lấy dữ liệu dưới dạng Dictionary
conn.row_factory = aiosqlite.Row
return conn
@asynccontextmanager
async def session(self):
"""
Context manager để tự động đóng Connection sau khi dùng xong.
Sử dụng:
async with sqlite_db.session() as conn:
await conn.execute(...)
"""
conn = await self.get_connection()
try:
yield conn
finally:
await conn.close()
async def fetch_all(self, query: str, params: tuple = ()) -> list[dict]:
"""Hàm tiện ích: Lọc toàn bộ danh sách, trả về list chứa các dict."""
async with self.session() as conn:
async with conn.execute(query, params) as cursor:
rows = await cursor.fetchall()
return [dict(row) for row in rows]
async def fetch_one(self, query: str, params: tuple = ()) -> Optional[dict]:
"""Hàm tiện ích: Lấy duy nhất 1 item đầu tiên, trả về dict."""
async with self.session() as conn:
async with conn.execute(query, params) as cursor:
row = await cursor.fetchone()
return dict(row) if row else None
async def execute(self, query: str, params: tuple = ()) -> int:
"""Hàm tiện ích: Chạy các truy vấn INSERT/UPDATE/DELETE và trả về số dòng bị ảnh hưởng."""
async with self.session() as conn:
async with conn.execute(query, params) as cursor:
await conn.commit()
return cursor.rowcount
# Export một Singleton instance để các file khác import vào dùng luôn
# Cách dùng: from common.sqlite_db import sqlite_db
sqlite_db = SQLiteDBManager()
backend/common/sqlite_mock.py 0 → 100644
View file @ 94484304
"""
sqlite_mock.py — Postgres Connection Interceptor (SQLite Fallback Layer)
KIẾN TRÚC:
- Module này giả lập các Psycopg2/Psycopg3 Connection/Cursor cho môi trường dev.
- MockCursor.execute() là SYNCHRONOUS (dùng sqlite3 trực tiếp).
- Lý do KHÔNG dùng aiosqlite ở đây: UltraDescriptionDB, DescFieldConfig và các
class helper khác gọi cursor.execute() theo cách ĐỒNG BỘ (classmethods thường).
Thay đổi sang async sẽ phá vỡ >24 method trong ultra_desc_db.py và nhiều hơn nữa.
LUỒNG ĐÚng:
- async API routes (FastAPI) dùng `sqlite_db.py` (aiosqlite) TRỰc TIẾP.
- Các helper class (UltraDescriptionDB, v.v.) gọi qua `pool_wrapper.get_pooled_connection_compat()`
→ trả về MockConnection → MockCursor SYNC.
"""
import logging
import re
import sqlite3
from contextlib import contextmanager
logger = logging.getLogger(__name__)
# ---------------------------------------------------------------------------
# DB Path (trỏ đến cùng file với sqlite_db.py)
# ---------------------------------------------------------------------------
from common.constants import SQLITE_DB_PATH
def _get_sync_conn() -> sqlite3.Connection:
"""Trả về kết nối sqlite3 ĐỒNG BỘ. Dùng cho MockCursor."""
conn = sqlite3.connect(SQLITE_DB_PATH)
conn.row_factory = sqlite3.Row
return conn
# ---------------------------------------------------------------------------
# SQL Dialect Translator: Postgres/StarRocks → SQLite
# ---------------------------------------------------------------------------
def translate_query(query: str) -> str:
"""
Dịch câu SQL đặc thù của Postgres/StarRocks sang dạng mà SQLite hiểu được.
"""
# 1. Thay thế Placeholder của Postgres (%s) thành của SQLite (?)
q = re.sub(r'(?<!%)%s(?!%)', '?', query)
# 2. Thay thế Postgres Schema (dashboard_canifa)
q = re.sub(r'"dashboard_canifa"\."([a-zA-Z0-9_]+)"', r'pg__dashboard_canifa__\1', q)
q = re.sub(r'dashboard_canifa\.([a-zA-Z0-9_]+)', r'pg__dashboard_canifa__\1', q)
# Thay thế Postgres Schema (canifa_chat)
q = re.sub(r'"canifa_chat"\."([a-zA-Z0-9_]+)"', r'pg__canifa_chat__\1', q)
q = re.sub(r'canifa_chat\.([a-zA-Z0-9_]+)', r'pg__canifa_chat__\1', q)
# Thay thế Postgres Schema (public)
q = re.sub(r'"public"\."([a-zA-Z0-9_]+)"', r'pg__public__\1', q)
q = re.sub(r'public\.([a-zA-Z0-9_]+)', r'pg__public__\1', q)
# 3. Thay thế cấu trúc của StarRocks
q = re.sub(
r'([a-zA-Z0-9_]+\.)?`?magento_product_dimension_with_text_embedding`?',
r'sr__test_db__magento_product_dimension_with_text_embedding',
q
)
# 4. Vá các ngữ pháp (Dialect) bị lệch pha
# ANY_VALUE(col) -> MAX(col)
q = re.sub(r'ANY_VALUE\s*\(', r'MAX(', q, flags=re.IGNORECASE)
# MAX_BY(col, score) -> MAX(col)
q = re.sub(r'MAX_BY\s*\(\s*([^,]+)\s*,\s*[^)]+\)', r'MAX(\1)', q, flags=re.IGNORECASE)
# IF(cond, true, false) -> IIF(cond, true, false)
q = re.sub(r'\bIF\s*\(', r'IIF(', q, flags=re.IGNORECASE)
# 5. Handle StarRocks vector similarity (dummy value)
q = re.sub(
r'approx_cosine_similarity\s*\(\s*vector\s*,\s*\[[^\]]+\]\s*\)',
r'0.5',
q, flags=re.IGNORECASE
)
# 6. Strip ORDER BY ... NULLS LAST
q = re.sub(r'NULLS\s+LAST', '', q, flags=re.IGNORECASE)
# 7. Strip PostgreSQL type casts (::jsonb, ::text, ::int, etc.)
q = re.sub(r'::[a-zA-Z_]+(?:\[\])?', '', q)
# 8. NOW() → CURRENT_TIMESTAMP
q = re.sub(r'\bNOW\s*\(\s*\)', 'CURRENT_TIMESTAMP', q, flags=re.IGNORECASE)
# 9. ILIKE → LIKE
q = re.sub(r'\bILIKE\b', 'LIKE', q, flags=re.IGNORECASE)
# 10. TRUE/FALSE literals → 1/0
q = re.sub(r'\bTRUE\b', '1', q, flags=re.IGNORECASE)
q = re.sub(r'\bFALSE\b', '0', q, flags=re.IGNORECASE)
# 11. ON CONFLICT DO NOTHING (SQLite supports this, keep as-is)
# 12. RETURNING id → SQLite không hỗ trợ RETURNING. Strip nó.
q = re.sub(r'\s+RETURNING\s+\S+', '', q, flags=re.IGNORECASE)
return q
# ---------------------------------------------------------------------------
# MockCursor — SYNCHRONOUS, dùng sqlite3 trực tiếp
# ---------------------------------------------------------------------------
class MockRow(tuple):
"""
Giả lập psycopg2 Row / aiomysql DictCursor kết hợp.
Hỗ trợ truy cập qua chỉ mục (row[0]) và qua tên cột (row['id']).
"""
def __new__(cls, cols, values):
return super().__new__(cls, values)
def __init__(self, cols, values):
self._dict = dict(zip(cols, values))
def __getitem__(self, key):
if isinstance(key, str):
return self._dict[key]
return super().__getitem__(key)
def keys(self):
return self._dict.keys()
def get(self, key, default=None):
return self._dict.get(key, default)
class MockCursor:
"""
Giả lập Psycopg2/3 Cursor, chạy ĐỒNG BỘ trên SQLite.
KHÔNG chuyển sang async để tránh phá vỡ:
- UltraDescriptionDB (24+ sync classmethods)
- DescFieldConfig (7+ sync classmethods)
- auth/models.py, agent/sql_agent/persistence.py, v.v.
"""
def __init__(self):
self._rows: list = []
self.rowcount: int = 0
self.description: list = []
self._last_insert_id: int | None = None
def execute(self, query: str, params: tuple = None):
"""Chạy truy vấn SQL — ĐỒNG BỘ, sqlite3 trực tiếp."""
if params is None:
params = ()
tq = translate_query(query)
logger.debug("[SQLITE MOCK] Query: %.120s", tq)
conn = _get_sync_conn()
try:
cur = conn.cursor()
cur.execute(tq, params)
is_select = tq.strip().upper().startswith("SELECT")
is_insert = tq.strip().upper().startswith("INSERT")
if is_select:
fetched = cur.fetchall()
if cur.description:
cols = [d[0] for d in cur.description]
self._rows = [MockRow(cols, row) for row in fetched]
self.description = [(d[0],) for d in cur.description]
else:
self._rows = []
self.rowcount = len(self._rows)
else:
conn.commit()
self.rowcount = cur.rowcount
if is_insert:
self._last_insert_id = cur.lastrowid
# Giả lập RETURNING id → trả về lastrowid dưới dạng hàng đầu tiên
self._rows = [MockRow(["id"], [cur.lastrowid])] if cur.lastrowid else []
else:
self._rows = []
except Exception as e:
logger.error("[SQLITE MOCK] Error: %s | Query: %.200s", e, tq)
self._rows = []
self.rowcount = 0
finally:
conn.close()
def executemany(self, query: str, params_list):
"""Chạy nhiều tham số — ĐỒNG BỘ."""
tq = translate_query(query)
conn = _get_sync_conn()
try:
cur = conn.cursor()
cur.executemany(tq, params_list)
conn.commit()
self.rowcount = cur.rowcount
self._rows = []
except Exception as e:
logger.error("[SQLITE MOCK] executemany error: %s", e)
finally:
conn.close()
def fetchall(self) -> list:
return self._rows
def fetchone(self) -> dict | tuple | None:
if self._rows:
return self._rows[0]
return None
def close(self):
pass
# Hỗ trợ `with cursor:` (sync context manager)
def __enter__(self):
return self
def __exit__(self, exc_type, exc_val, exc_tb):
self.close()
# ---------------------------------------------------------------------------
# MockConnection
# ---------------------------------------------------------------------------
class MockConnection:
"""
Giả lập psycopg2/3 Connection.
Trả về MockCursor đồng bộ khi gọi .cursor().
"""
def cursor(self) -> MockCursor:
return MockCursor()
def commit(self):
pass # Commit được thực hiện trong MockCursor.execute()
def rollback(self):
pass
def close(self):
pass
@contextmanager
def transaction(self):
yield self
# ---------------------------------------------------------------------------
# Factory functions — dùng bởi db_pool.py và pool_wrapper.py
# ---------------------------------------------------------------------------
@contextmanager
def get_mock_pg_conn():
"""
Context manager: mock kết nối PostgreSQL → SQLite.
Dùng bởi db_pool.CanifaDbPool.get_conn() khi USE_LOCAL_SQLITE=true.
"""
logger.info("🛡️ [SQLITE MOCK] Đánh chặn Postgres connection → SQLite local")
yield MockConnection()
def get_mock_pg_conn_compat() -> MockConnection:
"""
Trực tiếp trả về MockConnection (không context manager).
Dùng bởi pool_wrapper.get_pooled_connection_compat() khi USE_LOCAL_SQLITE=true.
"""
logger.info("🛡️ [SQLITE MOCK COMPAT] Đánh chặn Postgres connection → SQLite local")
return MockConnection()
backend/scratch/api_tester.py 0 → 100644
View file @ 94484304
import requests
import json
import time
BASE_URL = "http://localhost:5000/api/agent/chat-dev"
def test_query(query, description):
print(f"\n{'='*60}")
print(f"TEST: {description}")
print(f"QUERY: '{query}'")
print(f"{'='*60}")
payload = {
"user_query": query,
}
try:
start_time = time.time()
response = requests.post(BASE_URL, json=payload, timeout=60)
elapsed = time.time() - start_time
if response.status_code == 200:
data = response.json()
print(f"✅ Success ({elapsed:.2f}s)")
print(f"🤖 Response: {data.get('ai_response')}")
p_ids = data.get('product_ids', [])
print(f"📦 Product IDs ({len(p_ids)}): {', '.join(p_ids[:5])}{'...' if len(p_ids)>5 else ''}")
# Check diagnostics if available (some versions return it)
if "diagnostics" in data:
for d in data["diagnostics"]:
print(f" - {d.get('label')}: {d.get('elapsed_ms')}ms")
else:
print(f"❌ Error {response.status_code}: {response.text}")
except Exception as e:
print(f"⚠️ Connection failed: {e}")
if __name__ == "__main__":
print("🚀 Starting Canifa AI Stylist API Verification...")
# 1. Test Product Search (SQLite Mock Logic)
test_query("Tìm áo phông cho bé trai", "Product Search (SQLite Mock)")
# 2. Test Promotion (StarRocks Production)
test_query("Hôm nay có khuyến mãi gì hot không?", "Promotion Search (StarRocks)")
# 3. Test Knowledge (StarRocks Production)
test_query("Chính sách đổi trả của Canifa thế nào?", "Knowledge Search (StarRocks)")
# 4. Test Store (StarRocks Production)
test_query("Tìm cửa hàng ở Cầu Giấy", "Store Search (StarRocks)")
print("\n✅ Verification sequence completed.")
backend/server.py
View file @ 94484304
......@@ -122,4 +122,5 @@ if __name__ == "__main__":
reload=ENABLE_RELOAD,
reload_dirs=reload_dirs,
log_level="info",
loop="asyncio" if platform.system() == "Windows" else "auto"
)
backend/static/index.html
View file @ 94484304
This source diff could not be displayed because it is too large. You can view the blob instead.
backend/static/lead_flow/lead-flow.css 0 → 100644
View file @ 94484304
/* ═══════════════════════════════════════
CANIFA STYLIST PRO — Light Mode Edition
═══════════════════════════════════════ */
@import url('https://fonts.googleapis.com/css2?family=Inter:wght@300;400;500;600;700;800&family=Outfit:wght@400;600;700;800&display=swap');
:root {
/* Brand Colors */
--primary: #C8102E;
--primary-hover: #A30D25;
--primary-glow: rgba(200, 16, 46, 0.15);
/* Neutral Palette - Light Mode */
--bg-main: #F8F9FA;
--bg-surface: #FFFFFF;
--bg-card: #FFFFFF;
--bg-bubble-user: #C8102E;
--bg-bubble-ai: #F1F3F5;
--bg-input: #F8F9FA;
/* Accents */
--accent-blue: #007AFF;
--accent-gray: #E9ECEF;
/* Typography */
--text-main: #121212;
--text-secondary: #495057;
--text-muted: #ADB5BD;
--font-sans: 'Inter', sans-serif;
--font-display: 'Outfit', sans-serif;
/* Borders & Shadows */
--border: rgba(0, 0, 0, 0.08);
--border-bright: rgba(0, 0, 0, 0.12);
--shadow-sm: 0 2px 8px rgba(0,0,0,0.05);
--shadow-md: 0 8px 24px rgba(0,0,0,0.08);
--shadow-lg: 0 16px 48px rgba(0,0,0,0.12);
/* Transitions */
--ease: cubic-bezier(0.16, 1, 0.3, 1);
}
* { box-sizing: border-box; }
body {
font-family: var(--font-sans);
margin: 0; padding: 0;
background-color: var(--bg-main);
color: var(--text-main);
height: 100vh;
overflow: hidden;
-webkit-font-smoothing: antialiased;
}
/* ═══ NAVIGATION ═══ */
.nav-header {
background: rgba(255, 255, 255, 0.8);
backdrop-filter: blur(20px);
padding: 12px 32px;
display: flex;
justify-content: space-between;
align-items: center;
border-bottom: 1px solid var(--border);
z-index: 1000;
}
.nav-header h1 {
margin: 0;
font-family: var(--font-display);
font-weight: 800;
font-size: 1.2rem;
color: var(--primary);
letter-spacing: 2px;
}
.nav-links { display: flex; gap: 12px; }
.nav-links a {
color: var(--text-secondary);
text-decoration: none;
padding: 8px 16px;
border-radius: 10px;
background: var(--accent-gray);
transition: all 0.3s var(--ease);
font-size: 13px;
font-weight: 600;
}
.nav-links a:hover { background: #DEE2E6; transform: translateY(-2px); color: var(--text-main); }
.nav-links a.active { background: var(--primary); color: white; box-shadow: 0 4px 12px var(--primary-glow); }
/* ═══ MAIN LAYOUT ═══ */
.main-content {
height: calc(100vh - 64px);
max-width: 1600px;
margin: 0 auto;
padding: 24px;
display: flex;
gap: 24px;
}
/* CHAT CONTAINER */
.chat-container {
flex: 1;
display: flex;
flex-direction: column;
background: var(--bg-surface);
border-radius: 24px;
border: 1px solid var(--border);
box-shadow: var(--shadow-md);
overflow: hidden;
position: relative;
}
/* CONFIG BAR */
.config-bar {
padding: 16px 24px;
background: #F8F9FA;
border-bottom: 1px solid var(--border);
display: flex;
align-items: center;
gap: 16px;
flex-wrap: wrap;
}
.config-item { display: flex; align-items: center; gap: 8px; }
.config-item label { font-size: 11px; font-weight: 700; color: var(--text-muted); text-transform: uppercase; }
.config-item input {
background: white;
border: 1px solid var(--border);
border-radius: 8px;
color: var(--text-main);
padding: 6px 12px;
font-size: 12px;
outline: none;
}
.config-item input:focus { border-color: var(--primary); }
.btn-action {
padding: 8px 16px;
border-radius: 10px;
font-size: 12px;
font-weight: 700;
border: none;
cursor: pointer;
transition: all 0.3s var(--ease);
}
.btn-refresh { background: #007AFF; color: white; }
.btn-prompt { background: #FAB005; color: #000; }
.btn-clear { background: #FA5252; color: white; }
/* CHAT BOX */
.chat-box {
flex: 1;
overflow-y: auto;
padding: 32px;
display: flex;
flex-direction: column;
gap: 24px;
scroll-behavior: smooth;
background: #FFFFFF;
}
.msg-group { max-width: 85%; display: flex; flex-direction: column; }
.msg-group.user { align-self: flex-end; align-items: flex-end; }
.msg-group.bot { align-self: flex-start; align-items: flex-start; }
.sender-name { font-size: 11px; font-weight: 800; color: var(--text-muted); text-transform: uppercase; margin-bottom: 6px; letter-spacing: 1px; }
.bubble {
padding: 16px 20px;
border-radius: 20px;
font-size: 15px;
line-height: 1.6;
animation: fadeIn 0.4s var(--ease);
}
@keyframes fadeIn { from { opacity: 0; transform: translateY(10px); } to { opacity: 1; transform: translateY(0); } }
.bubble.user { background: var(--primary); color: white; border-bottom-right-radius: 4px; box-shadow: 0 4px 15px var(--primary-glow); }
.bubble.bot { background: var(--bg-bubble-ai); border: 1px solid var(--border-bright); color: var(--text-main); border-bottom-left-radius: 4px; }
.bubble .timestamp { display: block; font-size: 10px; opacity: 0.5; margin-top: 8px; text-align: right; }
/* PRODUCT CARDS */
.product-grid {
display: grid;
grid-template-columns: repeat(auto-fill, minmax(220px, 1fr));
gap: 20px;
margin-top: 20px;
width: 100%;
}
.p-card {
background: white;
border: 1px solid var(--border);
border-radius: 16px;
overflow: hidden;
transition: all 0.4s var(--ease);
display: flex;
flex-direction: column;
}
.p-card:hover { transform: translateY(-8px); border-color: var(--primary); box-shadow: var(--shadow-lg); }
.p-img-wrap { position: relative; height: 260px; overflow: hidden; background: #F8F9FA; }
.p-img { width: 100%; height: 100%; object-fit: cover; transition: transform 0.8s var(--ease); }
.p-card:hover .p-img { transform: scale(1.05); }
.p-badge {
position: absolute; top: 12px; left: 12px; background: var(--primary); color: white;
font-size: 10px; font-weight: 800; padding: 4px 10px; border-radius: 6px; z-index: 10;
}
.p-content { padding: 16px; flex: 1; display: flex; flex-direction: column; }
.p-sku { font-size: 11px; font-weight: 700; color: var(--text-muted); margin-bottom: 4px; }
.p-name { font-size: 14px; font-weight: 600; color: var(--text-main); margin-bottom: 12px; height: 40px; overflow: hidden; }
.p-price-row { display: flex; align-items: center; gap: 8px; margin-top: auto; margin-bottom: 16px; }
.p-price-sale { font-size: 18px; font-weight: 800; color: var(--primary); }
.p-price-old { font-size: 13px; color: var(--text-muted); text-decoration: line-through; }
.p-btn {
background: #F8F9FA; color: var(--text-main); border: 1px solid var(--border);
padding: 10px; border-radius: 10px; text-decoration: none; font-size: 12px; font-weight: 700;
text-align: center; transition: all 0.3s;
}
.p-btn:hover { background: var(--primary); border-color: var(--primary); color: white; }
/* INPUT BAR */
.input-bar {
padding: 24px;
background: var(--bg-surface);
border-top: 1px solid var(--border);
}
.input-inner {
background: #F8F9FA;
border: 1px solid var(--border-bright);
border-radius: 20px;
padding: 8px 8px 8px 24px;
display: flex;
gap: 12px;
align-items: center;
}
.input-inner input { flex: 1; background: transparent; border: none; color: var(--text-main); font-size: 16px; outline: none; }
.btn-send {
width: 48px; height: 48px; border-radius: 14px; background: var(--primary); color: white;
border: none; cursor: pointer; display: flex; align-items: center; justify-content: center;
}
/* PROMPT PANEL */
.prompt-panel {
width: 0; opacity: 0; visibility: hidden;
background: white; border-left: 1px solid var(--border);
display: flex; flex-direction: column; transition: all 0.5s var(--ease);
}
.prompt-panel.open { width: 500px; opacity: 1; visibility: visible; }
.prompt-inner { padding: 32px; display: flex; flex-direction: column; height: 100%; gap: 20px; }
.prompt-textarea {
flex: 1; background: #F8F9FA; color: #495057; border: 1px solid var(--border);
border-radius: 12px; padding: 20px; font-family: 'Consolas', monospace; font-size: 13px; line-height: 1.6;
}
backend/static/service/api.js 0 → 100644
View file @ 94484304
/**
* CANIFA STYLIST PRO - API Service
* Modularized API interactions
*/
const CanifaAPI = {
getDeviceId: () => {
let devId = localStorage.getItem('canifa_device_id');
if (!devId) {
devId = 'dev-' + Math.random().toString(36).substr(2, 8);
localStorage.setItem('canifa_device_id', devId);
}
return devId;
},
getHeaders: (deviceId, token = null) => {
const headers = {
'Content-Type': 'application/json',
'device_id': deviceId
};
if (token) {
headers['Authorization'] = 'Bearer ' + token;
}
return headers;
},
fetchHistory: async (deviceId, limit = 20) => {
try {
const res = await fetch(`/api/history/${deviceId}?limit=${limit}`, {
headers: CanifaAPI.getHeaders(deviceId)
});
return await res.json();
} catch (e) {
console.error('History fetch error:', e);
return { data: [] };
}
},
chat: async (query, deviceId, token = null, dbSource = 'starrocks') => {
const res = await fetch('/api/agent/chat-dev', {
method: 'POST',
headers: CanifaAPI.getHeaders(deviceId, token),
body: JSON.stringify({
user_query: query,
device_id: deviceId,
config: { configurable: { db_source: dbSource } }
})
});
return await res.json();
},
getPrompt: async (type) => {
const url = type === 'system' ? '/api/agent/system-prompt' : '/api/agent/tool-prompts/lead_search_tool.py';
const res = await fetch(url);
return await res.json();
},
savePrompt: async (type, content) => {
const url = type === 'system' ? '/api/agent/system-prompt' : '/api/agent/tool-prompts/lead_search_tool.py';
const res = await fetch(url, {
method: 'POST',
headers: { 'Content-Type': 'application/json' },
body: JSON.stringify({ content })
});
return await res.json();
},
archiveHistory: async (deviceId) => {
return await fetch('/api/history/archive', {
method: 'POST',
headers: CanifaAPI.getHeaders(deviceId)
});
}
};
window.CanifaAPI = CanifaAPI;
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment