Update utils/database.py

utils/database.py

@@ -337,7 +337,7 @@ def display_vector_store_info():
         st.error(traceback.format_exc())
         
 def initialize_qa_system(vector_store):
-    """Initialize QA system with structured responses."""
+    """Initialize QA system with clean response formatting."""
     try:
         llm = ChatOpenAI(
             temperature=0.5,
@@ -345,53 +345,45 @@ def initialize_qa_system(vector_store):
             api_key=os.environ.get("OPENAI_API_KEY")
         )
 
-        # Create retriever function
-        retriever = vector_store.as_retriever(search_kwargs={"k": 2})
-
-        # Create a template that encourages structured responses
+        # Create a template that enforces clean formatting
         prompt = ChatPromptTemplate.from_messages([
-            ("system", """You are a helpful assistant analyzing RFP documents to help companies reuse the knowledge based and be efficient with new RFP responses. 
-            When providing answers:
-            1. Structure your response clearly with sections
-            2. Use bullet points for key information
-            3. Always cite the source document and relevant sections
-            4. Provide a brief summary at the start
-            5. Be concise and clear"""),
-            MessagesPlaceholder(variable_name="chat_history"),
-            ("human", """{input}
-
-            Context from documents:
-            {context}
+            ("system", """You are a helpful assistant analyzing RFP documents. 
+            Format your responses in a clean, professional manner using markdown formatting:
             
-            Please provide a structured response with:
-            - Summary
-            - Key Points
-            - Source Details
-            """)
+            1. Start with a brief executive summary in plain text
+            2. Use proper markdown headers for sections (e.g., ## Key Points)
+            3. Use bullet points for lists
+            4. Include source attribution at the end in italics
+            5. Don't include any technical metadata or delimiters
+            6. Use bold for emphasis on important points
+            7. Keep the formatting clean and professional"""),
+            MessagesPlaceholder(variable_name="chat_history"),
+            ("human", "{input}\n\nContext: {context}")
         ])
 
-        def get_chat_history(inputs):
-            chat_history = inputs.get("chat_history", [])
-            if not isinstance(chat_history, list):
-                return []
-            return [msg for msg in chat_history if isinstance(msg, BaseMessage)]
-
-        def get_context_with_sources(inputs):
-            docs = retriever.get_relevant_documents(inputs["input"])
-            contexts = []
-            for doc in docs:
-                source = doc.metadata.get('source', 'Unknown source')
-                contexts.append(f"\nSource: {source}\nContent: {doc.page_content}")
-            return "\n---\n".join(contexts)
+        def format_response(response_text):
+            """Clean up the response formatting."""
+            # Remove technical metadata
+            if 'content=' in response_text:
+                response_text = response_text.split('content=')[1].split('response_metadata')[0]
+            
+            # Remove outer quotes if present
+            response_text = response_text.strip("'")
+            
+            # Remove escaped newlines and replace with actual newlines
+            response_text = response_text.replace('\\n', '\n')
+            
+            return response_text
 
         chain = (
             {
-                "context": get_context_with_sources,
-                "chat_history": get_chat_history,
+                "context": lambda x: get_context_with_sources(x["input"], retriever),
+                "chat_history": lambda x: x.get("chat_history", []),
                 "input": lambda x: x["input"]
             }
-            | prompt 
+            | prompt
             | llm
+            | format_response
         )
 
         return chain