Update app.py

app.py

@@ -98,6 +98,7 @@ else:
         persist_directory=persist_directory
     )
     print("Created new vector store and persisted embeddings.")
+
 # Create a BM25 retriever from the document splits
 bm25_retriever = BM25Retriever.from_documents(all_splits)
 
@@ -112,38 +113,19 @@ retriever = ensemble_retriever
 # Prepare Retrieval and Generation Chain
 # -------------------------------
 
-#### RETRIEVAL and GENERATION ####
-
-# Prompt
-prompt = hub.pull("rlm/rag-prompt")
-
-# LLM
-llm = ChatOpenAI(model_name="gpt-4o-mini", temperature=0)
-
-# Post-processing
-def format_docs(docs):
-    return "\n\n".join(doc.page_content for doc in docs)
-
-# Chain
-rag_chain = (
-    {"context": retriever | format_docs, "question": RunnablePassthrough()}
-    | prompt
-    | llm
-    | StrOutputParser()
-)
+# Update the system prompt to instruct the model on using provided context
 system_prompt = (
-    "You are the LA2050 Navigator, an AI-powered chatbot designed to help users explore organizations "
-    "and community initiatives within the Goldhirsh Foundation’s LA2050 Ideas Hub. "
-    "Your role is to provide personalized recommendations, guide users toward supporting these organizations "
-    "and initiatives, and answer relevant questions about the Goldhirsh Foundation, LA2050, and its projects. "
-    "When answering, include the full name of the organization, a brief (1-2 sentence) description, and a link to its website or social media "
-    "(as provided under the website column; please do not alter or normalize the URL). "
-    "Also include suggestions for engagement such as donating, volunteering, or learning more. "
-    "If a company's personal website is unavailable, navigate to the LA2050 URLs. "
+    "You are the LA2050 Navigator, an AI-powered chatbot designed to help users explore organizations and community initiatives within the Goldhirsh Foundation’s LA2050 Ideas Hub. "
+    "Your role is to provide personalized recommendations, guide users toward supporting these organizations and initiatives, and answer relevant questions about the Goldhirsh Foundation, LA2050, and its projects. "
+    "When answering, include the full name of the organization, a brief (1-2 sentence) description, and a link to its website or social media (as provided under the website column; please do not alter or normalize the URL). "
+    "Also include suggestions for engagement such as donating, volunteering, or learning more. If a company's personal website is unavailable, navigate to the LA2050 URLs. "
     "Prioritize nonprofit organizations awarded by the Goldhirsh Foundation (designated 'winner') and those with multiple proposal submissions. "
     "Use the data files as your primary source of information. If information is unavailable, acknowledge it and guide the user to relevant resources. "
-    "Maintain a polite, helpful, respectful, and enthusiastic tone at all times.\n\n{context}"
+    "Maintain a polite, helpful, respectful, and enthusiastic tone at all times. "
+    "If the user responds with a follow-up confirmation (e.g. 'yes') after a previous answer, please expand on that topic with additional information. "
+    "\n\n{context}"
 )
+
 prompt = ChatPromptTemplate.from_messages(
     [
         ("system", system_prompt),
@@ -151,10 +133,14 @@ prompt = ChatPromptTemplate.from_messages(
     ]
 )
 
-
-question_answer_chain = create_stuff_documents_chain(llm, prompt)
+# Build the chain that will combine documents with the prompt
+question_answer_chain = create_stuff_documents_chain(ChatOpenAI(model_name="gpt-4o-mini", temperature=0), prompt)
 rag_chain = create_retrieval_chain(retriever, question_answer_chain)
 
+# -------------------------------
+# Gradio Interface and Conversation Handling
+# -------------------------------
+
 green_theme = gr.themes.Base(
     primary_hue=gr.themes.Color(
         c50="#00A168",
@@ -186,23 +172,29 @@ green_theme = gr.themes.Base(
     button_secondary_text_color='#000000'
 )
 
-
-
-# integrate gradio with RAG logic
+# Here we update the conversation function so that each new query includes recent conversation history.
 def message_and_history(message, history):
+    # If no conversation exists yet, initialize with a welcome message.
     history = history or [{"role": "assistant", "content": "<b>LA2050 Navigator:</b><br> Welcome to the LA2050 ideas hub! How can I help you today?"}]
-    history.append({"role": "user", "content": message.get("text", "")})
-
+    user_text = message.get("text", "")
+    history.append({"role": "user", "content": user_text})
+    
     time.sleep(1)
 
-    user_input = message.get("text", "")
-    if not user_input:
+    # If the user did not provide any input, ask for a valid message.
+    if not user_text:
         history.append({"role": "assistant", "content": "<b>LA2050 Navigator:</b><br> Please enter a valid message."})
         yield history, history
         return
 
+    # Combine the most recent conversation turns to include as context (here, the last 6 messages)
+    conversation_context = "\n".join(
+        [f"{msg['role']}: {msg['content']}" for msg in history[-6:]]
+    )
+    chain_input = {"input": conversation_context}
+    
     try:
-        response = rag_chain.invoke({"input": user_input})
+        response = rag_chain.invoke(chain_input)
         answer = response["answer"]
     except Exception as e:
         answer = f"An error occurred: {e}"
@@ -210,40 +202,39 @@ def message_and_history(message, history):
     dynamic_message = {"role": "assistant", "content": "<b>LA2050 Navigator:</b><br> "}
     history.append(dynamic_message)
 
+    # Stream the answer character by character.
     for character in answer:
         dynamic_message["content"] += character
         yield history, history
 
+    # Finalize the answer.
     history[-1]["content"] = f"<b>LA2050 Navigator:</b><br> {answer}"
     yield history, history
 
 
-# set to light mode
+# Set Gradio to light mode via JavaScript
 js_func = """
 function refresh() {
     const url = new URL(window.location);
-
     if (url.searchParams.get('__theme') !== 'light') {
         url.searchParams.set('__theme', 'light');
         window.location.href = url.href;
     }
 }
 """
-css = """""
-    .chat-header {
+
+css = """
+.chat-header {
     text-color: #FFFFFF;
     text-align: center;
-    }
-    .gradio-container .prose .chat-header h1 {
+}
+.gradio-container .prose .chat-header h1 {
     color: #FFFFFF;
     text-align: center;
-    }
-
 }
 """
 
-# setup gradio interface
-
+# Setup Gradio interface with the custom theme, JS, and CSS
 with gr.Blocks(theme=green_theme, js=js_func, css=css) as block:
     gr.HTML('<div class="chat-header"><h1>LA2050 Navigator</h1></div>')
     
@@ -265,6 +256,7 @@ with gr.Blocks(theme=green_theme, js=js_func, css=css) as block:
         show_label=False
     )
     
+    # When a message is submitted, the function now sends the recent conversation history along with the new input.
     message.submit(
         message_and_history,
         inputs=[message, state],
@@ -274,3 +266,4 @@ with gr.Blocks(theme=green_theme, js=js_func, css=css) as block:
     )
     
 block.launch(debug=True, share=True)
+