Update main.py

main.py

@@ -10,6 +10,7 @@ import os
 from datetime import datetime
 from pymongo import MongoClient
 from huggingface_hub import InferenceClient
+
 from embedding_service import JinaClipEmbeddingService
 from qdrant_service import QdrantVectorService
 from advanced_rag import AdvancedRAG
@@ -124,7 +125,7 @@ class ChatRequest(BaseModel):
     message: str
     use_rag: bool = True
     top_k: int = 3
-    system_message: Optional[str] = """Bạn là trợ lý AI chuyên biệt cho hệ thống quản lý sự kiện và mạng xã hội. 
+    system_message: Optional[str] = """Bạn là trợ lý AI chuyên biệt cho hệ thống quản lý sự kiện và bán vé. 
 Vai trò của bạn là trả lời các câu hỏi CHÍNH XÁC dựa trên dữ liệu được cung cấp từ hệ thống.
 
 Quy tắc tuyệt đối:
@@ -685,29 +686,69 @@ async def chat(request: ChatRequest):
     try:
         # Retrieve context if RAG enabled
         context_used = []
+        rag_stats = None
+        
         if request.use_rag:
-            # Generate query embedding
-            query_embedding = embedding_service.encode_text(request.message)
-            
-            # Search in Qdrant
-            results = qdrant_service.search(
-                query_embedding=query_embedding,
-                limit=request.top_k,
-                score_threshold=0.5
-            )
-            context_used = results
-
-        # Build context text
-        context_text = ""
-        if context_used:
-            context_text = "\n\nRelevant Context:\n"
-            for i, doc in enumerate(context_used, 1):
-                doc_text = doc["metadata"].get("text", "")
-                confidence = doc["confidence"]
-                context_text += f"\n[{i}] (Confidence: {confidence:.2f})\n{doc_text}\n"
-
-            # Add context to system message
-            system_message = f"{request.system_message}\n{context_text}\n\nPlease use the above context to answer the user's question when relevant."
+            if request.use_advanced_rag:
+                # Use Advanced RAG Pipeline (Best Case 2025)
+                hf_client = None
+                if request.hf_token or hf_token:
+                    hf_client = InferenceClient(token=request.hf_token or hf_token)
+                
+                documents, stats = advanced_rag.hybrid_rag_pipeline(
+                    query=request.message,
+                    top_k=request.top_k,
+                    score_threshold=request.score_threshold,
+                    use_reranking=request.use_reranking,
+                    use_compression=request.use_compression,
+                    use_query_expansion=request.use_query_expansion,
+                    max_context_tokens=500,
+                    hf_client=hf_client
+                )
+                
+                # Convert to dict format
+                context_used = [
+                    {
+                        "id": doc.id,
+                        "confidence": doc.confidence,
+                        "metadata": doc.metadata
+                    }
+                    for doc in documents
+                ]
+                rag_stats = stats
+                
+                # Format context using Advanced RAG
+                context_text = advanced_rag.format_context_for_llm(documents)
+            else:
+                # Basic RAG (fallback)
+                query_embedding = embedding_service.encode_text(request.message)
+                results = qdrant_service.search(
+                    query_embedding=query_embedding,
+                    limit=request.top_k,
+                    score_threshold=request.score_threshold
+                )
+                context_used = results
+                
+                context_text = "\n\nRelevant Context:\n"
+                for i, doc in enumerate(context_used, 1):
+                    doc_text = doc["metadata"].get("text", "")
+                    if not doc_text:
+                        doc_text = " ".join(doc["metadata"].get("texts", []))
+                    confidence = doc["confidence"]
+                    context_text += f"\n[{i}] (Confidence: {confidence:.2f})\n{doc_text}\n"
+
+        # Build system message with context
+        if request.use_rag and context_used:
+            if request.use_advanced_rag:
+                # Use Advanced RAG prompt builder
+                system_message = advanced_rag.build_rag_prompt(
+                    query=request.message,
+                    context=context_text,
+                    system_message=request.system_message
+                )
+            else:
+                # Basic prompt
+                system_message = f"{request.system_message}\n{context_text}\n\nPlease use the above context to answer the user's question when relevant."
         else:
             system_message = request.system_message
 
@@ -771,7 +812,8 @@ Example:
         return ChatResponse(
             response=response,
             context_used=context_used,
-            timestamp=datetime.utcnow().isoformat()
+            timestamp=datetime.utcnow().isoformat(),
+            rag_stats=rag_stats
         )
 
     except Exception as e: