Update backend.py

backend.py

@@ -209,45 +209,44 @@ def get_embeddings_model():
 
 
 @st.cache_resource
-def initialize_qa_system(_vector_store):
+def initialize_qa_system(vector_store):
+    """Initialize QA system with proper chat handling."""
     try:
         llm = ChatOpenAI(
-            temperature=0,
-            model_name="gpt-4",  # Or another OpenAI model like "gpt-3.5-turbo"
-            api_key=os.environ.get("OPENAI_API_KEY"),
+            temperature=0.5,
+            model_name="gpt-4",
+            api_key=os.environ.get("OPENAI_API_KEY")
         )
 
-        # Define the prompt template (ADD agent_scratchpad)
-        prompt = ChatPromptTemplate.from_messages(
-            [
-                ("system", "You are a helpful assistant"),
-                MessagesPlaceholder(variable_name="chat_history"),
-                ("human", "{input}"),
-                MessagesPlaceholder(variable_name="agent_scratchpad"),
-            ]
+        # Create a more basic prompt template
+        prompt = ChatPromptTemplate.from_messages([
+            ("system", """You are an expert consultant specializing in analyzing Request for Proposal (RFP) documents. 
+            Your goal is to provide clear, accurate responses based on the provided context. 
+            Start with a direct answer and organize additional details under relevant headers."""),
+            ("human", "{input}")
+        ])
+
+        # Create the retriever chain
+        retriever = vector_store.as_retriever(
+            search_type="similarity",
+            search_kwargs={"k": 3}
         )
 
-        # Define the tools
-        tools = [
-            Tool(
-                name="Search",
-                func=_vector_store.as_retriever(
-                    search_kwargs={"k": 2}
-                ).get_relevant_documents,
-                description="useful for when you need to answer questions about the documents you have been uploaded. Input should be a fully formed question.",
-            )
-        ]
-
-        # Create the agent and executor
-        agent = create_openai_tools_agent(llm=llm, tools=tools, prompt=prompt)
-        agent_executor = AgentExecutor(
-            agent=agent,
-            tools=tools,
-            verbose=True,
-            memory=ConversationBufferMemory(memory_key="chat_history"),
+        chain = (
+            {
+                "input": RunnablePassthrough()
+            }
+            | {"input": lambda x: x["input"], "docs": retriever}
+            | {
+                "input": lambda x: x["input"],
+                "context": lambda x: "\n\n".join([doc.page_content for doc in x["docs"]])
+            }
+            | prompt
+            | llm
         )
 
-        return agent_executor  # Return the agent executor
+        return chain
+
     except Exception as e:
         st.error(f"Error initializing QA system: {e}")
         return None