feat: optimize filter performance and rename query to description -...

feat: optimize filter performance and rename query to description - Pre-compile regex patterns for filter optimization - Rename SearchItem.query to SearchItem.description for semantic search - Add váy đầm, váy bé gái to product_map - Fix price format (remove decimal) - Tune LIMIT values for vector search

backend/agent/graph.py

@@ -147,16 +147,30 @@ 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)."""
+    from .prompt import get_last_modified
+    
+    current_prompt_mtime = get_last_modified()
+    
     # 1. New Instance if Empty
     if _instance[0] is None:
         _instance[0] = CANIFAGraph(config, llm, tools)
+        _instance[0].prompt_mtime = current_prompt_mtime
         logger.info(f"✨ Graph Created: {_instance[0].config.model_name}")
         return _instance[0]
 
-    # 2. Check for Config Changes (e.g. Model Switch)
-    if config and config.model_name != _instance[0].config.model_name:
-        logger.info(f"🔄 Model Switch: {_instance[0].config.model_name} -> {config.model_name}")
+    # 2. Check for Config Changes (Model Switch OR Prompt Update)
+    is_model_changed = config and config.model_name != _instance[0].config.model_name
+    is_prompt_changed = current_prompt_mtime != getattr(_instance[0], "prompt_mtime", 0.0)
+
+    if is_model_changed or is_prompt_changed:
+        change_reason = []
+        if is_model_changed: change_reason.append(f"Model ({_instance[0].config.model_name}->{config.model_name})")
+        if is_prompt_changed: change_reason.append("Prompt File Updated")
+        
+        logger.info(f"🔄 Rebuilding Graph due to: {', '.join(change_reason)}")
+        
         _instance[0] = CANIFAGraph(config, llm, tools)
+        _instance[0].prompt_mtime = current_prompt_mtime
         return _instance[0]
 
     return _instance[0]

backend/agent/mock_controller.py

@@ -248,7 +248,19 @@ async def mock_chat_controller(
             price_min=None,
             price_max=None,
             action="search",
+            # Metadata fields - all required with None default
+            gender_by_product=None,
+            age_by_product=None,
+            product_name=None,
+            style=None,
+            master_color=None,
+            season=None,
+            material_group=None,
+            fitting=None,
+            form_neckline=None,
+            form_sleeve=None,
         )
+
         result_json = await data_retrieval_tool.ainvoke({"searches": [search_item]})
         result = json.loads(result_json)
         all_product_ids = result.get("results", [{}])[0].get("products", [])

backend/agent/prompt.py

@@ -37,3 +37,13 @@ Hôm nay: {date_str}
 
 KHÔNG BAO GIỜ BỊA ĐẶT. TRẢ LỜI NGẮN GỌN.
 """
+
+
+def get_last_modified() -> float:
+    """Trả về timestamp lần sửa cuối cùng của file system_prompt.txt."""
+    try:
+        if os.path.exists(PROMPT_FILE_PATH):
+            return os.path.getmtime(PROMPT_FILE_PATH)
+    except Exception:
+        pass
+    return 0.0
\ No newline at end of file

backend/agent/structured_models.py

+"""
+Structured Output Models for CANIFA Chatbot.
+Forces LLM to return strictly structured JSON responses.
+Uses OpenAI's native Structured Outputs feature via LangChain.
+"""
+
+from pydantic import BaseModel, Field
+
+
+class UserInsight(BaseModel):
+    """6-layer User Insight structure as defined in system prompt."""
+    
+    USER: str = Field(
+        default="Chưa rõ.",
+        description="Thông tin người chat: Giới tính, Người lớn/Trẻ em, Style/Gu"
+    )
+    TARGET: str = Field(
+        default="Chưa rõ.",
+        description="Đối tượng thụ hưởng: Quan hệ, Giới tính, Người lớn/Trẻ em, Style/Gu"
+    )
+    GOAL: str = Field(
+        default="Chưa rõ.",
+        description="Mục tiêu mua sắm: Sản phẩm + Dịp sử dụng (Occasion)"
+    )
+    CONSTRAINS: str = Field(
+        default="Chưa có.",
+        description="Ràng buộc cứng: Budget, Size, Màu, Chất liệu, TRÁNH XA/GHÉT..."
+    )
+    LATEST_PRODUCT_INTEREST: str = Field(
+        default="Chưa có",
+        description="Sản phẩm vừa mới hỏi/xem gần nhất"
+    )
+    NEXT: str = Field(
+        default="Cần hỏi thêm thông tin.",
+        description="Chiến lược tiếp theo của bot"
+    )
+    SUMMARY_HISTORY: str = Field(
+        default="",
+        description="Tóm tắt lịch sử chat quan trọng"
+    )
+
+    class Config:
+        extra = "forbid"  # Required for OpenAI strict mode
+
+
+class ChatResponse(BaseModel):
+    """
+    Structured response from CANIFA AI Stylist.
+    This model enforces the exact JSON schema the LLM must follow.
+    """
+    
+    ai_response: str = Field(
+        description="Câu trả lời cho khách hàng. Phải ngắn gọn, thảo mai, nhắc SKU bằng [SKU]."
+    )
+    product_ids: list[str] = Field(
+        default_factory=list,
+        description="Danh sách mã SKU sản phẩm được nhắc đến (chỉ string, không object)"
+    )
+    user_insight: UserInsight = Field(
+        default_factory=UserInsight,
+        description="Insight 6 tầng về khách hàng, cập nhật sau mỗi turn"
+    )
+
+    class Config:
+        extra = "forbid"  # Required for OpenAI strict mode

backend/agent/tools/product_search_helpers.py

@@ -87,6 +87,7 @@ async def build_starrocks_query(params, query_vector: list[float] | None = None)
     if magento_code:
         logger.info(f"🎯 [CODE SEARCH] Direct search by code: {magento_code}")
         
+
         sql = """
         SELECT 
             internal_ref_code,
@@ -94,9 +95,10 @@ async def build_starrocks_query(params, query_vector: list[float] | None = None)
             sale_price,
             original_price,
             discount_amount,
+            product_line_vn,
+            product_line_en,
             1.0 as max_score
         FROM shared_source.magento_product_dimension_with_text_embedding
-        WHERE (magento_ref_code = %s OR internal_ref_code = %s)
         """
         return sql, [magento_code, magento_code]
 
@@ -105,7 +107,7 @@ async def build_starrocks_query(params, query_vector: list[float] | None = None)
     # ============================================================
     logger.info("🚀 [HYDE RETRIEVER] Starting semantic vector search...")
     
-    query_text = getattr(params, "query", None)
+    query_text = getattr(params, "description", None)
     if query_text and query_vector is None:
         emb_start = time.time()
         query_vector = await create_embedding_async(query_text)
@@ -142,7 +144,7 @@ async def build_starrocks_query(params, query_vector: list[float] | None = None)
             approx_cosine_similarity(vector, {v_str}) as similarity_score
         FROM shared_source.magento_product_dimension_with_text_embedding
         ORDER BY similarity_score DESC
-        LIMIT 100
+        LIMIT 200
     )
     SELECT 
         internal_ref_code,
@@ -155,7 +157,7 @@ async def build_starrocks_query(params, query_vector: list[float] | None = None)
     WHERE 1=1 {where_filter}
     GROUP BY internal_ref_code
     ORDER BY max_score DESC
-    LIMIT 20
+    LIMIT 70
     """
     
     # Return sql and params (params only contains filter values now, not the vector)

backend/api/mock_api_route.py

@@ -111,15 +111,27 @@ async def mock_db_search(req: MockDBRequest):
         logger.info("📍 Data Retrieval Tool called")
         start_time = time.time()
 
-        # Xây dựng SearchItem từ request
+        # Xây dựng SearchItem từ request - include all required fields
         search_item = SearchItem(
             query=req.query or "sản phẩm",
             magento_ref_code=req.magento_ref_code,
             price_min=req.price_min,
             price_max=req.price_max,
             action="search",
+            # Metadata fields - all required with None default
+            gender_by_product=None,
+            age_by_product=None,
+            product_name=None,
+            style=None,
+            master_color=None,
+            season=None,
+            material_group=None,
+            fitting=None,
+            form_neckline=None,
+            form_sleeve=None,
         )
 
+
         logger.info(f"🔧 Search params: {search_item.dict(exclude_none=True)}")
 
         # Gọi data_retrieval_tool THẬT với retry
@@ -161,15 +173,27 @@ async def mock_retriever_db(req: MockRetrieverRequest):
         logger.info(f"📍 Retriever DB started: {req.user_query}")
         start_time = time.time()
 
-        # Xây dựng SearchItem từ request
+        # Xây dựng SearchItem từ request - include all required fields
         search_item = SearchItem(
             query=req.user_query,
             magento_ref_code=req.magento_ref_code,
             price_min=req.price_min,
             price_max=req.price_max,
             action="search",
+            # Metadata fields - all required with None default
+            gender_by_product=None,
+            age_by_product=None,
+            product_name=None,
+            style=None,
+            master_color=None,
+            season=None,
+            material_group=None,
+            fitting=None,
+            form_neckline=None,
+            form_sleeve=None,
         )
 
+
         logger.info(f"🔧 Retriever params: {search_item.dict(exclude_none=True)}")
 
         # Gọi data_retrieval_tool THẬT (embedding + vector search) với retry