Update app.py

app.py

@@ -571,31 +571,46 @@ def qa_llm():
     return qa
 
 def process_answer(user_question, full_text):
-    """Generate an answer to the user’s question or summarize the PDF content."""
+    """Generate an answer to the user’s question based on the extracted text from the PDF."""
     try:
         logging.info("Processing user question")
+
+        # Set up the retriever with the PDF content (this could be your embedded database or a direct retrieval from full_text)
+        embeddings = SentenceTransformerEmbeddings(model_name="all-MiniLM-L6-v2")
         
-        # Check if the question is related to summarization
-        if "summarize" in user_question.lower() or "summary" in user_question.lower():
-            tailored_prompt = f"""
-            Please provide a summary of the following content extracted from the PDF:
-            {full_text}
-            """
-        else:
-            # Regular Q&A with context from the uploaded PDF
-            tailored_prompt = f"""
-            You are an expert chatbot designed to assist with any topic, providing accurate and detailed answers based on the provided PDFs. 
-            Your goal is to deliver the most relevant information and resources based on the question asked. 
-            User question: {user_question}
-            Content from the uploaded document: {full_text}
-            """
-
-        # Pass the tailored prompt to the question-answering chain (QA) system
-        qa = qa_llm()  # Call your QA LLM setup
-        generated_text = qa({"query": tailored_prompt})
+        # Use Chroma for document storage and retrieval if you’re storing documents in a vector store
+        db = Chroma(persist_directory=persist_directory, embedding_function=embeddings)
+        retriever = db.as_retriever()  # Set up the retriever to use Chroma database
+        
+        # Here we're just adding the full_text as a document for simplicity
+        db.add_documents([full_text])
+
+        # Set up the language model pipeline (assuming you already have a pipeline set up)
+        llm = llm_pipeline()
+
+        # Construct the retrieval chain using the retriever and LLM
+        qa_chain = RetrievalQA.from_chain_type(
+            llm=llm,
+            chain_type="stuff",
+            retriever=retriever,
+            return_source_documents=True
+        )
+        
+        # Create a tailored prompt for the question (providing context to the chatbot)
+        tailored_prompt = f"""
+        You are a helpful RAG-based chatbot designed to assist with answering questions from any uploaded document. 
+        You should answer the question using relevant information from the provided PDF text.
+        Please provide a clear, informative answer based on the document content.
+        User question: {user_question}
+        """
+
+        # Generate the answer using the retrieval-augmented generation model
+        generated_text = qa_chain({"query": tailored_prompt})
+
+        # Extract the generated answer
         answer = generated_text['result']
 
-        # If the answer contains certain fallback phrases, return a default message
+        # If the answer is empty or not very informative, provide a fallback message
         if "not provide" in answer or "no information" in answer:
             return "The document does not provide sufficient information to answer your question."
 
@@ -604,7 +619,7 @@ def process_answer(user_question, full_text):
 
     except Exception as e:
         logging.error(f"Error during answer generation: {str(e)}")
-        return "Error processing the question."
+        return "Sorry, I encountered an issue while processing your question."
 
 
 # Streamlit UI Setup