Update app.py

app.py

@@ -124,43 +124,35 @@ async def health_check():
 
 @app.post("/query", response_model=QueryResponse, tags=["Query"])
 async def query_rag(request: QueryRequest):
-    """
-    Query the RAG system
-    """
     global last_qa_context
-    
+
     if not rag_system:
-        raise HTTPException(
-            status_code=503,
-            detail="RAG system not initialized. Check logs for errors."
-        )
-    
-    try:
-        # Build prompt using previous Q/A if available
-        if last_qa_context:
-            prompt = f"{last_qa_context}\n\n{request.question}"
-        else:
-            prompt = request.question
+        raise HTTPException(status_code=503, detail="RAG system not initialized.")
 
-        # Query RAG system
-        result = rag_system.ask(prompt)
+    try:
+        # Pass ONLY the question to retrieval
+        result = rag_system.ask(
+            request.question,
+            context=last_qa_context
+        )
 
-        # Save current Q/A as context for next turn
+        # Save lightweight context for next turn
         last_qa_context = (
-            f"Question: {request.question}\n"
-            f"Answer: {result['answer']}"
+            f"Q: {request.question}\n"
+            f"A: {result['answer'][:500]}"  # truncate for safety
         )
-        
+
         return QueryResponse(
-            answer=result['answer'],
-            images=result['images'],
-            texts=result['texts'],
+            answer=result["answer"],
+            images=result["images"],
+            texts=result["texts"],
             question=request.question
         )
-        
+
     except Exception as e:
         logger.error(f"Error processing query: {str(e)}")
-        raise HTTPException(status_code=500, detail=f"Error processing query: {str(e)}")
+        raise HTTPException(status_code=500, detail=str(e))
+
 
 if __name__ == "__main__":
     import uvicorn