Update app.py

app.py

@@ -1,5 +1,6 @@
 import os
 import sys
+import re
 import gradio as gr
 from langchain_openai import ChatOpenAI
 from langgraph.prebuilt import create_react_agent
@@ -13,39 +14,74 @@ NEBIUS_API_KEY = os.getenv("NEBIUS_API_KEY")
 NEBIUS_BASE_URL = "https://api.studio.nebius.ai/v1/"
 MODEL_NAME = "meta-llama/Meta-Llama-3.1-70B-Instruct"
 
-# --- Agent System Prompt ---
+# --- Advanced System Prompt with Structured Output ---
 SYSTEM_PROMPT = """You are a 'Vibe Coding' Python Tutor. 
-Your goal is not just to talk, but to DO.
-1. When a user asks to learn a concept, create a python file illustrating it.
-2. RUN the file to show them the output.
-3. If there is an error, debug it by reading the file and fixing it.
-4. Always explain your reasoning briefly before executing tools.
+Your goal is to teach by DOING and then providing resources.
 
-You have access to a local filesystem. Use 'write_file' to create examples and 'run_python_script' to execute them.
+STRUCTURE OF YOUR RESPONSE:
+1. **The Lesson**: Explain the concept and run code using your tools ('write_file', 'run_python_script').
+2. **Context**: Use 'list_directory' to see what the student is working on.
+
+CRITICAL: You must end EVERY response with these exact separators to populate the student's dashboard:
+
+---SECTION: VIDEOS---
+(List 2-3 YouTube search queries or URLs relevant to the topic)
+
+---SECTION: ARTICLES---
+(List 2-3 documentation links or course names, e.g., RealPython, FreeCodeCamp)
+
+---SECTION: QUIZ---
+(Create 1 short multiple-choice question to test what you just taught)
 """
 
-async def run_tutor(user_message, chat_history):
+def parse_agent_response(full_text):
+    """Splits the single LLM response into 4 UI components."""
+    # Default content if sections are missing
+    chat_content = full_text
+    videos = "Ask a coding question to get video recommendations!"
+    articles = "Ask a coding question to get reading resources!"
+    quiz = "Ask a coding question to take a quiz!"
+
+    # Robust parsing using split
+    try:
+        if "---SECTION: VIDEOS---" in full_text:
+            parts = full_text.split("---SECTION: VIDEOS---")
+            chat_content = parts[0].strip()
+            remainder = parts[1]
+            
+            if "---SECTION: ARTICLES---" in remainder:
+                v_parts = remainder.split("---SECTION: ARTICLES---")
+                videos = v_parts[0].strip()
+                remainder = v_parts[1]
+                
+                if "---SECTION: QUIZ---" in remainder:
+                    a_parts = remainder.split("---SECTION: QUIZ---")
+                    articles = a_parts[0].strip()
+                    quiz = a_parts[1].strip()
+                else:
+                    articles = remainder.strip()
+            else:
+                videos = remainder.strip()
+    except Exception as e:
+        print(f"Parsing error: {e}")
+
+    return chat_content, videos, articles, quiz
+
+async def run_tutor_dashboard(user_message, chat_history):
     """
-    Main function to run the agent loop. 
-    It connects to the local MCP server for every request to ensure fresh context.
+    Main function to run the agent loop and return 4 outputs.
     """
-    
-    # 1. Define Server Parameters
     server_params = StdioServerParameters(
         command=sys.executable, 
         args=["server.py"], 
         env=os.environ.copy()
     )
 
-    # 2. Connect to MCP Server & Load Tools
     async with stdio_client(server_params) as (read, write):
         async with ClientSession(read, write) as session:
             await session.initialize()
-            
-            # Convert MCP tools to LangChain tools
             tools = await load_mcp_tools(session)
             
-            # 3. Initialize Nebius LLM
             llm = ChatOpenAI(
                 api_key=NEBIUS_API_KEY,
                 base_url=NEBIUS_BASE_URL,
@@ -53,43 +89,81 @@ async def run_tutor(user_message, chat_history):
                 temperature=0.7
             )
 
-            # 4. Create Agent
             agent_executor = create_react_agent(llm, tools, state_modifier=SYSTEM_PROMPT)
-
-            # 5. Execute
+            
+            # Run Agent
             inputs = {"messages": [HumanMessage(content=user_message)]}
             response = await agent_executor.ainvoke(inputs)
+            final_text = response["messages"][-1].content
             
-            # 6. Extract the final response text
-            return response["messages"][-1].content
+            return parse_agent_response(final_text)
 
-# --- Gradio UI (Universal Fix) ---
-with gr.Blocks(title="AI Python Tutor (MCP Powered)") as demo:
-    gr.Markdown("# 🐍 Vibe Coding Tutor")
+# --- Gradio Dashboard UI ---
+with gr.Blocks(title="AI Python Tutor (MCP Dashboard)", theme=gr.themes.Soft()) as demo:
+    gr.Markdown("# 🚀 Vibe Coding Academy")
     gr.Markdown("Powered by **Nebius** (Llama 3.1) & **MCP** (Local Filesystem Access)")
     
-    # REMOVED: type="messages" (This fixes the TypeError)
-    chatbot = gr.Chatbot(height=600) 
-    msg = gr.Textbox(placeholder="E.g., Teach me how to use Python decorators with a working example.")
-    
+    with gr.Row():
+        # Left Column: Chat & Input
+        with gr.Column(scale=2):
+            chatbot = gr.Chatbot(height=500, label="Tutor Chat", type="messages")
+            
+            # BOX 1: Learning Request (Input)
+            msg = gr.Textbox(
+                label="1. What do you want to learn?", 
+                placeholder="E.g., How do Python dictionaries work?",
+                lines=2
+            )
+            submit_btn = gr.Button("Start Learning", variant="primary")
+
+        # Right Column: Resources Dashboard
+        with gr.Column(scale=1):
+            # BOX 2: Videos
+            video_box = gr.Markdown(
+                value="### 📺 Recommended Videos\n*Waiting for topic...*", 
+                label="2. Video References"
+            )
+            
+            # BOX 3: Articles/Courses
+            article_box = gr.Markdown(
+                value="### 📚 Articles & Courses\n*Waiting for topic...*", 
+                label="3. Articles & Courses"
+            )
+            
+            # BOX 4: Quiz
+            quiz_box = gr.Markdown(
+                value="### 🧠 Quick Quiz\n*Waiting for topic...*", 
+                label="4. Knowledge Check"
+            )
+
+    # --- Interaction Logic ---
     async def user_turn(user_message, history):
-        # Universal format: Append a list of [user_msg, None]
-        # This works on Gradio 3, 4, and 5
-        return "", history + [[user_message, None]]
+        return "", history + [{"role": "user", "content": user_message}]
 
     async def bot_turn(history):
-        # Get the last user message (it's the first element of the last tuple)
-        last_message = history[-1][0]
+        last_message = history[-1]["content"]
         
-        # Run the agent
-        response_text = await run_tutor(last_message, [])
+        # Get all 4 outputs from the agent
+        chat_text, video_text, article_text, quiz_text = await run_tutor_dashboard(last_message, [])
         
-        # Update the last tuple with the bot response
-        history[-1][1] = response_text
-        return history
+        # Update Chatbot history
+        history.append({"role": "assistant", "content": chat_text})
+        
+        # Return all 4 updates
+        return history, video_text, article_text, quiz_text
 
-    msg.submit(user_turn, [msg, chatbot], [msg, chatbot]).then(
-        bot_turn, [chatbot], [chatbot]
+    # Wire up the inputs and outputs
+    # Notice we now output to [chatbot, video_box, article_box, quiz_box]
+    submit_btn.click(
+        user_turn, [msg, chatbot], [msg, chatbot]
+    ).then(
+        bot_turn, [chatbot], [chatbot, video_box, article_box, quiz_box]
+    )
+    
+    msg.submit(
+        user_turn, [msg, chatbot], [msg, chatbot]
+    ).then(
+        bot_turn, [chatbot], [chatbot, video_box, article_box, quiz_box]
     )
 
 # --- Launch ---