app.py

from faster_whisper import WhisperModel
import os
import requests
import json
from langchain.tools import tool
from typing import TypedDict
from langgraph.graph import StateGraph
import gradio as gr
from pydub import AudioSegment
import tempfile

MISTRAL_API_KEY = os.getenv("MISTRAL_API_KEY")
MODEL = "mistral-large-latest"

class State(TypedDict):
    file_path: str
    transcript: str
    summary_json: dict
    final_output: str

def convert_to_wav(input_path: str) -> str:
    if input_path.lower().endswith(".wav"):
        return input_path  # No conversion needed
    else:
        audio = AudioSegment.from_file(input_path)
        temp_wav_file = tempfile.NamedTemporaryFile(delete=False, suffix=".wav")
        audio.export(temp_wav_file.name, format="wav")
        return temp_wav_file.name

def transcribe_audio(state: State) -> State:
    model = WhisperModel("base", compute_type="float32")
    segments, _ = model.transcribe(state["file_path"], beam_size=5, language="en")

    transcript = " ".join(segment.text.strip() for segment in segments)
    return {**state, "transcript": transcript}

def summarize_transcript(state: State) -> State:
    prompt = f"""
You are an AI assistant summarizing meeting transcripts.
Extract the following:
1. Attendees (Names only, if possible)
2. Key Discussion Points (bullet format)
3. Action Items (with owner and deadline, if mentioned)
4. Decisions Made (if any)
Respond in this format:
{{
  "attendees": [...],
  "key_points": [...],
  "action_items": [...],
  "decisions": [...]
}}
Transcript:
\"\"\"{state['transcript']}\"\"\"
"""
    res = requests.post(
        "https://api.mistral.ai/v1/chat/completions",
        headers={
            "Authorization": f"Bearer {MISTRAL_API_KEY}",
            "Content-Type": "application/json"
        },
        json={
            "model": MODEL,
            "messages": [{"role": "user", "content": prompt}],
            "temperature": 0.3
        }
    )
    content = res.json()['choices'][0]['message']['content']
    return {**state, "summary_json": eval(content)}

def format_output(state: State) -> State:
    summary = state["summary_json"]
    formatted = "📌 Attendees:\n" + "\n".join(f"- {name}" for name in summary["attendees"]) + "\n\n"
    formatted += "💬 Key Points:\n" + "\n".join(f"- {pt}" for pt in summary["key_points"]) + "\n\n"
    formatted += "✅ Action Items:\n" + "\n".join(f"- {item}" for item in summary["action_items"]) + "\n\n"
    formatted += "📎 Decisions:\n" + "\n".join(f"- {d}" for d in summary["decisions"])
    return {**state, "final_output": formatted}

# Graph setup
builder = StateGraph(State)
builder.add_node("transcribe_audio", transcribe_audio)
builder.add_node("summarize_transcript", summarize_transcript)
builder.add_node("format_output", format_output)

builder.set_entry_point("transcribe_audio")
builder.add_edge("transcribe_audio", "summarize_transcript")
builder.add_edge("summarize_transcript", "format_output")

graph = builder.compile()

def process_meeting(file):
    print(f"Received file: {file}")
    wav_path = convert_to_wav(file)
    state = {"file_path": wav_path}
    print(f"State initialized: {state}")

    try:
        final_state = graph.invoke(state)
        print(f"Final State: {final_state}")
        return final_state["final_output"]
    except Exception as e:
        print(f"Error: {e}")
        return str(e)

with gr.Blocks() as demo:
    gr.Markdown("### 🤖 MeetRecap - AI-Powered Meeting Summarizer")
    gr.Markdown("""
    Upload your audio file (e.g., `.wav`, `.mp3`, `.m4a`) or record it. You'll receive a structured meeting summary including Attendees, Key Points, Action Items, and Decisions.
    """)

    with gr.Row():
        with gr.Column():
            audio_input = gr.Audio(type="filepath", label="🎙️ Upload or Record Audio")
            submit_btn = gr.Button("🔍 Summarize")
            clear_btn = gr.Button("🧹 Clear")
        with gr.Column():
            summary_output = gr.Textbox(label="📝 Meeting Summary", lines=20, max_lines=40)

    submit_btn.click(fn=process_meeting, inputs=audio_input, outputs=summary_output)
    clear_btn.click(lambda: (None, ""), outputs=[audio_input, summary_output])

demo.launch(share=True)