app.py

import gradio as gr
import os
import json
import uuid
import threading
import time
import re

from openai import OpenAI
from dotenv import load_dotenv
from realtime_transcriber import WebSocketClient, connections, WEBSOCKET_URI, WEBSOCKET_HEADERS

# ------------------ Load API Key ------------------
load_dotenv()
OPENAI_API_KEY = os.getenv("OPENAI_API_KEY")
ASSISTANT_ID = os.getenv("ASSISTANT_ID")

if not OPENAI_API_KEY or not ASSISTANT_ID:
    raise ValueError("Missing OPENAI_API_KEY or ASSISTANT_ID in environment variables")

client = OpenAI(api_key=OPENAI_API_KEY)

# ------------------ Chat Logic ------------------
session_threads = {}

def reset_session():
    session_id = str(uuid.uuid4())
    thread = client.beta.threads.create()
    session_threads[session_id] = thread.id
    return session_id

def process_chat(message, history, session_id):
    thread_id = session_threads.get(session_id)
    if not thread_id:
        thread_id = client.beta.threads.create().id
        session_threads[session_id] = thread_id

    client.beta.threads.messages.create(
        thread_id=thread_id,
        role="user",
        content=message
    )

    run = client.beta.threads.runs.create(
        thread_id=thread_id,
        assistant_id=ASSISTANT_ID
    )

    while True:
        run_status = client.beta.threads.runs.retrieve(
            thread_id=thread_id,
            run_id=run.id
        )
        if run_status.status == "completed":
            break
        time.sleep(1)

    messages = client.beta.threads.messages.list(thread_id=thread_id)
    assistant_response = "⚠️ Assistant did not respond."
    for msg in reversed(messages.data):
        if msg.role == "assistant":
            assistant_response = msg.content[0].text.value
            break

    return assistant_response

def extract_image_url(text):
    match = re.search(
        r'https://raw\.githubusercontent\.com/AndrewLORTech/surgical-pathology-manual/main/[\w\-/]*\.png',
        text
    )
    return match.group(0) if match else None

# ------------------ Transcription Logic ------------------
def create_websocket_client():
    client_id = str(uuid.uuid4())
    connections[client_id] = WebSocketClient(WEBSOCKET_URI, WEBSOCKET_HEADERS, client_id)
    threading.Thread(target=connections[client_id].run, daemon=True).start()
    return client_id

def clear_transcript(client_id):
    if client_id in connections:
        connections[client_id].transcript = ""
    return ""

def send_audio_chunk(audio, client_id):
    if client_id not in connections:
        return "Initializing connection..."
    sr, y = audio
    connections[client_id].enqueue_audio_chunk(sr, y)
    return connections[client_id].transcript

# ------------------ Gradio App ------------------
with gr.Blocks(theme=gr.themes.Soft()) as demo:
    gr.Markdown("# 🧠 Document AI + 🎙️ Voice Assistant")

    session_id = gr.State(value=reset_session())
    client_id = gr.State()
    image_url = gr.State(value=None)

    with gr.Row():
        with gr.Column(scale=1):
            image_display = gr.Image(label="📑 Extracted Document Image", show_label=True, height=400)
        with gr.Column(scale=2):
            chatbot = gr.ChatInterface(
                fn=lambda message, history, session_id: process_chat(message, history, session_id),
                additional_inputs=[session_id],
                examples=[
                    ["What does clause 3.2 mean?"],
                    ["Summarize the timeline from the image."]
                ],
                title="💬 Document Assistant"
            )

    # Inject logic to extract image when assistant replies
    def handle_reply_and_update_image(message, history, session_id):
        response = process_chat(message, history, session_id)
        url = extract_image_url(response)
        return response, url

    chatbot.fn = lambda message, history, session_id: handle_reply_and_update_image(message, history, session_id)[0]
    chatbot.chatbot.change(
        fn=lambda m, h, s: handle_reply_and_update_image(m, h, s)[1],
        inputs=[chatbot.input, chatbot.chatbot, session_id],
        outputs=image_display
    )

    # ------------------ Voice Transcription ------------------
    gr.Markdown("## 🎙️ Realtime Voice Transcription")

    with gr.Row():
        transcript_box = gr.Textbox(label="Live Transcript", lines=7, interactive=False, autoscroll=True)

    with gr.Row():
        mic_input = gr.Audio(streaming=True)
        clear_button = gr.Button("Clear Transcript")

    mic_input.stream(fn=send_audio_chunk, inputs=[mic_input, client_id], outputs=transcript_box)
    clear_button.click(fn=clear_transcript, inputs=[client_id], outputs=transcript_box)
    demo.load(fn=create_websocket_client, outputs=client_id)

demo.launch()