app.py

import gradio as gr
import requests
import time
import json
import os
import datetime
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

### SET YOUR ASSEMBLYAI API KEY
ASSEMBLYAI_API_KEY = os.getenv("ASSEMBLYAI_API_KEY", "your_assemblyai_api_key")
headers = {"authorization": ASSEMBLYAI_API_KEY}
notes_file = "notes.json"

### LOAD LLM
model_id = "IlmaJiyadh/phi3-4k-ft"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    device_map="auto",
    torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
    trust_remote_code=True
)

### TRANSCRIBE AUDIO WITH ASSEMBLYAI
def transcribe(audio_path):
    with open(audio_path, 'rb') as f:
        upload_res = requests.post("https://api.assemblyai.com/v2/upload", headers=headers, files={"file": f})
    audio_url = upload_res.json()["upload_url"]

    transcript_res = requests.post("https://api.assemblyai.com/v2/transcript", json={"audio_url": audio_url}, headers=headers)
    transcript_id = transcript_res.json()["id"]

    while True:
        poll = requests.get(f"https://api.assemblyai.com/v2/transcript/{transcript_id}", headers=headers).json()
        if poll['status'] == 'completed':
            return poll['text']
        elif poll['status'] == 'error':
            return f"Transcription failed: {poll['error']}"
        time.sleep(2)

### SUMMARIZE USING LLM
def summarize(text):
    prompt = f"Below is a lecture transcript. Take lecture notes in bullet points.\n\nInput:\n{text}\n\nSummary:\n"
    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
    outputs = model.generate(**inputs, max_new_tokens=200, temperature=0.7)
    return tokenizer.decode(outputs[0], skip_special_tokens=True)

### SAVE TO JSON
def save_note(date, transcript, summary):
    data = {"date": date, "transcript": transcript, "summary": summary}
    if os.path.exists(notes_file):
        with open(notes_file, "r") as f:
            all_notes = json.load(f)
    else:
        all_notes = []
    all_notes.append(data)
    with open(notes_file, "w") as f:
        json.dump(all_notes, f, indent=2)

### SEARCH NOTES
def search_notes(query):
    if not os.path.exists(notes_file):
        return "No notes available yet."
    with open(notes_file, "r") as f:
        notes = json.load(f)
    results = [n for n in notes if query.lower() in n['summary'].lower() or query.lower() in n['transcript'].lower()]
    if not results:
        return "No matching notes found."
    return "\n\n".join([f"📅 {n['date']}\n{n['summary']}" for n in results])

### FULL PIPELINE
def full_pipeline(audio):
    if audio is None:
        return "No audio provided", "", ""
    
    transcript = transcribe(audio)
    summary = summarize(transcript)
    date_str = str(datetime.date.today())
    save_note(date_str, transcript, summary)
    return transcript, summary, f"✅ Lecture saved for {date_str}"

### BUILD GRADIO UI
with gr.Blocks() as demo:
    gr.Markdown("# 🎓 Lecture Assistant (Audio → Summary + Search)")
    with gr.Row():
        with gr.Column():
            #audio_input = gr.Audio(source="microphone", type="filepath", label="🎙️ Record Audio")
            audio_input = gr.Audio(type="filepath", label="🎙️ Record Audio")
            submit_btn = gr.Button("Transcribe & Summarize")
            transcript_output = gr.Textbox(label="📄 Transcript")
            summary_output = gr.Textbox(label="📝 Summary")
            save_status = gr.Textbox(label="💾 Save Status")

        with gr.Column():
            search_query = gr.Textbox(label="🔍 Search Notes")
            search_btn = gr.Button("Search")
            search_output = gr.Textbox(label="Results")

    submit_btn.click(fn=full_pipeline, inputs=audio_input, outputs=[transcript_output, summary_output, save_status])
    search_btn.click(fn=search_notes, inputs=search_query, outputs=search_output)

demo.launch()