workapp.py

# app.py
import re  # <--- CRITICAL: Needed for cleaning URLs
import gradio as gr
import os
import time
import json
from openai import OpenAI
from dotenv import load_dotenv
from mcp import ClientSession, StdioServerParameters
from mcp.client.stdio import stdio_client
from elevenlabs.client import ElevenLabs

load_dotenv()

# --- CONFIGURATION ---
client = OpenAI(
    base_url="https://api.hyperbolic.xyz/v1",
    api_key=os.getenv("HYPERBOLIC_API_KEY"), 
)

MODEL_NAME = "meta-llama/Meta-Llama-3.1-70B-Instruct"

# app.py
samclient = OpenAI(
    # SambaNova is FAST. Use it for the conversational parts.
    base_url="https://api.sambanova.ai/v1", 
    api_key=os.getenv("SAMBANOVA_API_KEY"), 
)
# Use their Llama 3.1 405B (High IQ) or 70B (High Speed)
samMODEL_NAME = "Meta-Llama-3.1-70B-Instruct"


eleven = ElevenLabs(api_key=os.getenv("ELEVENLABS_API_KEY"))
VOICE_ID = "JBFqnCBsd6RMkjVDRZzb" 

SYSTEM_PROMPT = """
You are Anato-Mitra, a strict Indian Medical College Professor.
- If the student asks a question, ALWAYS use the 'search_anatomy_diagrams' tool.
- Be concise and strict.
- The diagram will be shown automatically, so you just need to explain the clinical significance.
"""

# --- 1. THE BRAIN (Agent Logic) ---
async def run_agent(user_message, history):
    # Connect to MCP Server
    server_params = StdioServerParameters(
        command="uv", 
        args=["run", "viva_server.py"], 
    )

    async with stdio_client(server_params) as (read, write):
        async with ClientSession(read, write) as session:
            await session.initialize()
            tools_list = await session.list_tools()
            
            openai_tools = [{
                "type": "function",
                "function": {
                    "name": t.name,
                    "description": t.description,
                    "parameters": t.inputSchema
                }
            } for t in tools_list.tools]

            messages = [{"role": "system", "content": SYSTEM_PROMPT}]
            messages.append({"role": "user", "content": user_message})

            print("🧠 Thinking...")
            response = client.chat.completions.create(
                model=MODEL_NAME,
                messages=messages,
                tools=openai_tools,
                tool_choice="auto"
            )

            final_response = response.choices[0].message.content or ""
            tool_image_markdown = "" 

            # Handle Tool Calls
            if response.choices[0].message.tool_calls:
                tool_call = response.choices[0].message.tool_calls[0]
                fn_name = tool_call.function.name
                fn_args = tool_call.function.arguments
                
                print(f"🔧 Tool Call: {fn_name}")
                
                args_dict = json.loads(fn_args)
                result = await session.call_tool(fn_name, arguments=args_dict)
                tool_output = result.content[0].text
                
                # Capture the image to display later
                tool_image_markdown = f"\n\n---\n**Reference Diagram:**\n{tool_output}"
                
                # Context Injection
                messages.append({
                    "role": "user", 
                    "content": f"SYSTEM DATA: {tool_output}\n\nAnswer the student."
                })
                
                final_response_obj = client.chat.completions.create(
                    model=MODEL_NAME,
                    messages=messages
                )
                final_response = final_response_obj.choices[0].message.content

            # Force show the diagram if one was found
            if tool_image_markdown:
                final_response += tool_image_markdown

            # Note: We do NOT generate audio here anymore.
            return final_response

# --- 2. THE VOICE (Clean Audio Generator) ---
def generate_audio(chat_history):
    """
    Generates audio for the LAST message.
    Cleans URLs and Markdown before sending to ElevenLabs.
    """
    if not chat_history:
        return None
        
    # Get the last message from the bot
    last_bot_message = chat_history[-1]['content']
    
    # --- CLEANING STEP (Regex) ---
    # 1. Remove Markdown Images like ![Diagram](...)
    clean_text = re.sub(r'!\[.*?\]\(.*?\)', '', last_bot_message)
    
    # 2. Remove raw URLs like https://...
    clean_text = re.sub(r'https?://\S+', '', clean_text)
    
    # 3. Cleanup specific phrases
    clean_text = clean_text.replace("Reference Diagram:", "").strip()
    
    print(f"🗣️ Speaking (Cleaned): {clean_text[:50]}...") 

    try:
        audio_generator = eleven.text_to_speech.convert(
            text=clean_text,
            voice_id=VOICE_ID,
            model_id="eleven_multilingual_v2"
        )
        
        audio_path = f"voice_{int(time.time())}.mp3"
        with open(audio_path, "wb") as f:
            for chunk in audio_generator:
                f.write(chunk)
        return audio_path
    except Exception as e:
        print(f"Audio Error: {e}")
        return None

# --- 3. THE UI ---
with gr.Blocks(theme=gr.themes.Soft()) as demo:
    gr.Markdown("# 🩻 Anato-Mitra: VIVA Companion  -Upper Limb & Thorax Anatomy")
    
    with gr.Row():
        with gr.Column(scale=2):
            chatbot = gr.Chatbot(type="messages", height=500)
            msg = gr.Textbox(placeholder="Ask: 'Show me the Circle of Willis'")
            
        with gr.Column(scale=1):
            gr.Markdown("### 🔊 VIVA Examiner Controls")
            audio_out = gr.Audio(label="Audio Output", type="filepath")
            # This button calls the generate_audio function manually
            btn_speak = gr.Button("🔊 Speak Response (Save Credits)", variant="primary")

    async def respond(message, chat_history):
        bot_message = await run_agent(message, chat_history)
        chat_history.append({"role": "user", "content": message})
        chat_history.append({"role": "assistant", "content": bot_message})
        return "", chat_history

    msg.submit(respond, [msg, chatbot], [msg, chatbot])
    
    # Connect the button to the Clean Audio Function
    btn_speak.click(generate_audio, inputs=chatbot, outputs=audio_out)

if __name__ == "__main__":
    demo.launch(share=True, auth=("waheed", "jcet123"))