AI YouTube Question API

This repository contains a FastAPI application that performs Retrieval-Augmented Generation (RAG) using YouTube video content. The API takes a YouTube URL and a query, processes the video content to extract relevant information, and provides an AI-generated response based on the query.

Test The API

The API is deployed on Vercel, but the youtube-transcript-api does not work in the deployed version.

The youtube-transcript-api library should function locally and you can also use the pytube library to get more metadata info.

Test the API using preloaded videos from this Next.js app Test The API.

How It Works

1. Text Extraction: The YouTube video's transcript is extracted and split into chunks.
2. Vector Store: The text chunks are embedded using OpenAI embeddings and stored in Pinecone.
3. Querying: When a query is made, the relevant text chunks are retrieved from Pinecone.
4. Response Generation: The query is augmented with retrieved contexts and sent to OpenRouter for generating a detailed response.

Prerequisites

• Python 3.8+
• A Pinecone API key
• A OpenAI API key
• A OpenRouter API key

API Endpoints

POST /perform_rag

• youtube_url: The URL of the YouTube video to process.
• query: The question to answer.

Response:

• Success: Returns the AI-generated answer based on the provided query.

Middleware

Request Middleware (Optional)

• Logs requests, enforces rate limiting, and measures response processing time. It is currently commented out but can be activated if needed.

CORS Configuration

• The application is configured to allow requests from https://youtube-chatbot.vercel.app. Modify this in the CORSMiddleware settings to your frontend host.