YouTube RAG Azure web app - Cloud project

pip install flask
pip install yt-dlp
pip install openai-whisper
pip install llama-index-core
pip install llama-index-readers-file
pip install llama-index-llms-ollama
pip install llama-index-embeddings-huggingface

docker pull ollama/ollama

docker run --name ollama -d -p 11411:11411 ollama/ollama

docker exec -it ollama ollama run llama3

docker commit ollama ollama/ollama-llama3

import os
import subprocess
import uuid
from flask import Flask, request, render_template
import yt_dlp
import whisper

# LlamaIndex / Ollama imports
from llama_index.core import VectorStoreIndex, Document, Settings
from llama_index.embeddings.huggingface import HuggingFaceEmbedding
from llama_index.llms.ollama import Ollama

app = Flask(__name__)

############################
#   Configure LlamaIndex   #
############################
# Adjust the endpoint to reflect the address of your Docker container. 
# If you are running the container locally as a test, by default Ollama's REST API is on port 11411.
# If Docker is on the same machine, 'localhost' should be fine.
# If running in the Azure Container Instance, remove localhost and add the IP address of the container.
OLLAMA_ENDPOINT = "http://localhost:11434"

# Set up embeddings and LLM once (shared config)
Settings.embed_model = HuggingFaceEmbedding(model_name="BAAI/bge-base-en-v1.5")
Settings.llm = Ollama(model="llama3", base_url=OLLAMA_ENDPOINT, request_timeout=360.0)

@app.route("/", methods=["GET", "POST"])
def index():
    transcription = None
    summary = None

    if request.method == "POST":
        youtube_url = request.form.get("url", "").strip()
        if youtube_url:
            # 1. Generate unique filenames for temp files
            unique_id = str(uuid.uuid4())
            video_filename = f"video_{unique_id}.mp4"
            audio_filename = f"audio_{unique_id}.wav"

            # 2. Download the audio from YouTube using yt_dlp
            ydl_opts = {
                'outtmpl': video_filename,
                'format': 'bestaudio/best',
                'quiet': True,
                'cookiesfrombrowser': ('firefox',)
            }

            try:
                with yt_dlp.YoutubeDL(ydl_opts) as ydl:
                    ydl.download([youtube_url])
            except Exception as e:
                transcription = f"Error downloading video: {e}"
                return render_template("index.html", transcription=transcription)

            # 3. Extract audio using ffmpeg
            cmd = [
                "ffmpeg", "-y",
                "-i", video_filename,
                "-ar", "16000",
                "-ac", "1",
                audio_filename
            ]
            try:
                subprocess.run(cmd, check=True, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
            except subprocess.CalledProcessError as e:
                transcription = f"Error extracting audio: {e}"
                if os.path.exists(video_filename):
                    os.remove(video_filename)
                return render_template("index.html", transcription=transcription)

            # 4. Transcribe audio using Whisper
            try:
                whisper_model = whisper.load_model("base")  # or 'small', 'large', etc.
                result = whisper_model.transcribe(audio_filename)
                transcription = result["text"].strip()
            except Exception as e:
                transcription = f"Error transcribing audio: {e}"

            # Clean up temporary files
            if os.path.exists(video_filename):
                os.remove(video_filename)
            if os.path.exists(audio_filename):
                os.remove(audio_filename)

            # 5. Build an in-memory LlamaIndex with the transcription
            if transcription and not transcription.startswith("Error"):
                try:
                    doc = Document(text=transcription)
                    index = VectorStoreIndex.from_documents([doc])
                    query_engine = index.as_query_engine()

                    # 6. Hard-coded query to summarize the main subject of the transcript
                    user_query = "Summarize the main subject of the video"
                    response = query_engine.query(user_query)
                    summary = str(response)
                except Exception as e:
                    summary = f"Error with RAG query: {e}"

    return render_template("index.html", transcription=transcription, summary=summary)

if __name__ == "__main__":
    app.run(debug=True)

<!DOCTYPE html>
<html lang="en">
<head>
    <meta charset="UTF-8">
    <meta name="viewport" content="width=device-width">
    <title>Video Transcription and Summarization</title>
</head>
<body>
    <h1>Video Transcription and Summarization</h1>
    <form method="POST" action="/" style="margin-bottom:20px;">
        <label for="url">YouTube URL:</label><br>
        <input type="text" id="url" name="url" style="width:300px;" required><br><br>
        <label for="question">Your Question:</label>
        <input type="text" id="question" name="question"><br><br>
        <button type="submit">Submit</button>
    </form>

    <h2>Transcription:</h2>
    <p>{{ transcription }}</p>

    <h2>Summary/Answer:</h2>
    <p>{{ summary }}</p>
</body>
</html>