YouTube RAG - Video Chat Assistant 🎥💬

A powerful application that transforms YouTube videos into interactive conversations using Simple Retrieval-Augmented Generation RAG Architecture. Ask questions about YouTube video and get precise answers based on its content.

Features

• Video Processing: Extract and analyze content from YouTube video with available transcripts
• Multi-Language Support: Process videos with transcripts in multiple languages
• Simple RAG Pipeline: Accurately retrieve and contextualize relevant information from videos
• Interactive Chat: Ask questions about the video content and receive detailed answers
• Conversation History: Save and export your conversation history in JSON or Markdown format
• Modern UI: Clean and intuitive Streamlit interface

Technical Overview

The project consists of two main components:

1. FastAPI Backend: The core RAG system that processes videos and answers questions
2. Streamlit Frontend: User-friendly interface for interacting with the RAG system

RAG Architecture

The application uses a the Simple RAG pipeline:

1. Video Processing:

   • Extract transcript from YouTube video
   • Split transcript into meaningful chunks
   • Generate embeddings using Cohere's embed-english-v3.0 model
   • Store embeddings in Pinecone vector database

2. Question Answering:

   • Process user question through MultiQueryRetriever
   • Retrieve most relevant transcript chunks from Pinecone
   • Generate comprehensive answer using Google's Gemini 2.5 Flash model

Installation

Prerequisites

• Python 3.10+
• API keys for:
  • Pinecone
  • Cohere
  • Google AI (Gemini)

Setup

1. Clone the repository

2. Create and activate a virtual environment:

   python -m venv youtube_rag_env
   # On Windows
   .\youtube_rag_env\Scripts\activate
   # On Unix or MacOS
   source youtube_rag_env/bin/activate

3. Install the required packages:

   pip install -r requirements.txt

4. Create a .env file in the root directory with your API keys:

   GOOGLE_API_KEY="your_google_api_key"
   COHERE_API_KEY="your_cohere_api_key"
   PINECONE_API_KEY="your_pinecone_api_key"
   

Running the Application

1. Start the FastAPI backend:

   uvicorn youtube_rag_api:app --reload

2. In a separate terminal, start the Streamlit frontend:

   streamlit run streamlit_video_chat.py

3. Open your browser and navigate to http://localhost:8501

API Endpoints

Endpoint	Method	Purpose	Request	Response
/	GET	Health check	None	Status message
/process_video	POST	Process video transcript	video_url, language	Processing status
/chat	POST	Answer questions about a video	video_id, question	Answer with sources

Usage

1. Paste a YouTube URL in the input field
2. Select the language of the video's transcript
3. Click "Process Video" to extract and analyze the content
4. Once processing is complete, ask questions about the video in the chat interface
5. Use the suggested questions or type your own
6. Download your conversation history as JSON or Markdown from the sidebar

Project Structure

• youtube_rag_api.py - FastAPI backend implementing the RAG system
• streamlit_video_chat.py - Streamlit frontend for user interaction
• requirements.txt - Required Python packages
• notebooks/YouTubeRAG.ipynb - Development notebook with detailed explanations
• img/technical_architecture.png - Technical architecture diagram

Environment Variables

The application requires the following environment variables:

• GOOGLE_API_KEY - API key for Google's Gemini model
• COHERE_API_KEY - API key for Cohere's embedding model
• PINECONE_API_KEY - API key for Pinecone vector database

Limitations

• Only works with YouTube videos that have available transcripts
• Quality of answers depends on the quality and accuracy of video transcripts
• Processing very long videos may take more time and resources

Technologies Used

• Backend: FastAPI, LangChain, Pinecone
• Frontend: Streamlit
• AI/ML:
  • Cohere Embeddings (embed-english-v3.0)
  • Google Gemini 2.5 Flash
• Data: YouTube Transcript API
• Data Validation: Pydantic