Update app.py

app.py

@@ -104,101 +104,168 @@ retriever = ensemble_retriever
 # Prepare Retrieval and Generation Chain
 # -------------------------------
 
-# Pull the prompt from the hub; ensure that the prompt exists at the specified location
+#### RETRIEVAL and GENERATION ####
+
+# Prompt
 prompt = hub.pull("rlm/rag-prompt")
 
-# Initialize the language model (adjust the model name as needed)
+# LLM
 llm = ChatOpenAI(model_name="gpt-4o-mini", temperature=0)
 
-# Create the document chain (the "stuff" chain that combines retrieved documents)
-question_answer_chain = create_stuff_documents_chain(llm, prompt)
+# 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()
+)
+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."
+    "Adhere to these instructions when interacting with the user:"
+    "Start each session by introducing yourself and explaining your purpose before the user inputs any message."
+    "Provide tailored recommendations at a time by default unless the user requests a different number."
+    "Include: the full name of the organization; a brief (1-2 sentence) description; link to the organization personal websites/social"
+    "in the data files; suggestions for how the user can engage, such as donating, volunteering, or learning more."
+    "In your recommendations, prioritize nonprofit organizations awarded by the Goldhirsh Foundation (designated “winner”), as well as organizations with multiple proposal submissions."
+    "Use the data files as your primary source of information when answering questions. If information is unavailable in the files, acknowledge this and guide the user to relevant resources."
+    "Maintain a polite, helpful, respectful, and enthusiastic tone at all times to create a positive user experience."
+    "Politely redirect the user back to relevant topics if they ask unrelated questions, keeping the conversation aligned with your purpose."
+    "Use the following information about the data files to help you:"
+    "The “ideas” are community initiatives submitted by organizations to apply for grants and support from the Goldhirsh Foundation."
+    "There are currently no winners for the 2024 data, although this information can be found on the LA2050 Ideas Hub website. "
+    "Only projects labeled as “winner” in the Project Ranking column should be identified as winning projects."
+    "\n\n{context}"
+)
+prompt = ChatPromptTemplate.from_messages(
+    [
+        ("system", system_prompt),
+        ("human", "{input}"),
+    ]
+)
 
-# Create the retrieval augmented generation (RAG) chain using the retriever and document chain
+
+question_answer_chain = create_stuff_documents_chain(llm, prompt)
 rag_chain = create_retrieval_chain(retriever, question_answer_chain)
 
+green_theme = gr.themes.Base(
+    primary_hue=gr.themes.Color(
+        c50="#00A168",
+        c100="#57B485",
+        c200="#D7ECE0",
+        c300="#FFFFFF",
+        c400="#EAE9E9",
+        c500="#000000",
+        c600="#3A905E",
+        c700="#2A774A",
+        c800="#1A5E36",
+        c900="#0A4512",
+        c950="#052A08"
+    ),
+    font=[gr.themes.GoogleFont('Space Grotesk'), 'ui-sans-serif', 'system-ui', 'sans-serif']
+).set(
+    body_background_fill='#00A168',
+    body_text_color='#000000',
+    background_fill_primary='#FFFFFF',
+    background_fill_secondary='#FFFFFF',
+    border_color_accent='#57B485',
+    border_color_accent_subdued='#EAE9E9',
+    color_accent='#57B485',
+    color_accent_soft='#D7ECE0',
+    checkbox_background_color='#FFFFFF',
+    button_primary_background_fill='#57B485',
+    button_primary_background_fill_hover='#3A905E',
+    button_secondary_background_fill='#D7ECE0',
+    button_secondary_text_color='#000000'
+)
 
-# -------------------------------
-# Define the Chat Callback Function
-# -------------------------------
 
-def message_and_history(user_message, history):
-    """
-    Processes the user input, performs retrieval and generation,
-    and updates the conversation history.
-    """
-    # Initialize history if empty
-    if not history:
-        history = [{"role": "assistant", "content": "<b>LA2050 Navigator:</b><br> Welcome to the LA2050 ideas hub! How can I help you today?"}]
-    
-    # Append the user's message to history
-    history.append({"role": "user", "content": user_message})
-    
-    # Simulate a brief delay (optional)
+
+# integrate gradio with RAG logic
+def message_and_history(message, history):
+    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", "")})
+
     time.sleep(1)
-    
-    # If the input is empty, return an error message
-    if not user_message.strip():
+
+    user_input = message.get("text", "")
+    if not user_input:
         history.append({"role": "assistant", "content": "<b>LA2050 Navigator:</b><br> Please enter a valid message."})
         yield history, history
         return
 
     try:
-        # Invoke the RAG chain with the user's input
-        response = rag_chain.invoke({"input": user_message})
+        response = rag_chain.invoke({"input": user_input})
         answer = response["answer"]
     except Exception as e:
         answer = f"An error occurred: {e}"
 
-    # Prepare a dynamic response that simulates streaming text
     dynamic_message = {"role": "assistant", "content": "<b>LA2050 Navigator:</b><br> "}
     history.append(dynamic_message)
 
-    # Stream the answer character by character (this loop yields intermediate updates)
     for character in answer:
         dynamic_message["content"] += character
         yield history, history
 
-    # Finalize the answer and yield the final history
     history[-1]["content"] = f"<b>LA2050 Navigator:</b><br> {answer}"
     yield history, history
 
 
-# -------------------------------
-# Set Up the Gradio Interface
-# -------------------------------
-
-# Define a custom green theme for the interface
-green_theme = gr.themes.Base(
-    primary_hue=gr.themes.Color(
-        c50="#00A168", c100="#57B485", c200="#D7ECE0", c300="#FFFFFF",
-        c400="#EAE9E9", c500="#000000", c600="#3A905E", c700="#2A774A",
-        c800="#1A5E36", c900="#0A4512", c950="#052A08"
-    )
-)
+# set to light mode
+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 {
+    text-color: #FFFFFF;
+    text-align: center;
+    }
+    .gradio-container .prose .chat-header h1 {
+    color: #FFFFFF;
+    text-align: center;
+    }
+
+}
+"""
+
+# setup gradio interface
+with gr.Blocks(theme=green_theme, js=js_func, css=css) as block:
 
-with gr.Blocks(theme=green_theme) as block:
     gr.HTML('<div class="chat-header"><h1>LA2050 Navigator</h1></div>')
-    
-    # Initialize the chatbot with a welcome message
-    chatbot = gr.Chatbot(
-        value=[{"role": "assistant", "content": "<b>LA2050 Navigator:</b><br> Welcome to the LA2050 ideas hub! How can I help you today?"}],
-        type="messages"
-    )
-    
-    # Use a Gradio State to keep track of the conversation history
+
+    chatbot = gr.Chatbot( value=
+                         [{"role": "assistant", "content": "<b>LA2050 Navigator:</b><br> Welcome to the LA2050 ideas hub! How can I help you today?"}],
+                            type="messages",
+                            bubble_full_width=False)
+
     state = gr.State([])
-    
-    # Textbox for user input
-    user_input_box = gr.Textbox(placeholder="Type a message", scale=3, show_label=False)
-    
-    # When the textbox is submitted, run the callback function
-    user_input_box.submit(
+
+    message = gr.MultimodalTextbox(
+        interactive=True,
+        file_count="multiple",
+        placeholder="Type a message",
+        label="",
+        elem_classes="custom-textbox",
+        scale=3,
+        show_label=False)
+
+    message.submit(
         message_and_history,
-        inputs=[user_input_box, state],
+        inputs=[message, state],
         outputs=[chatbot, state]
     ).then(
-        lambda: "", inputs=[], outputs=user_input_box
+        lambda: "", inputs=[], outputs=message
     )
 
 block.launch(debug=True, share=True)
-