VideoRAG

AI-powered multi-video understanding and retrieval system for extracting insights from YouTube videos through semantic search, summarization, comparison, and study workflows.

---

Overview

VideoRAG enables users to analyze and interact with long-form video content such as lectures, documentaries, podcasts, crime cases, medical discussions, and educational videos.

Instead of manually searching through hours of content, users can upload one or multiple YouTube videos, generate structured summaries, ask contextual questions, compare explanations across videos, and retrieve the exact timestamps where answers appear.

The system combines transcript extraction, semantic chunking, vector retrieval, and LLM-powered reasoning to create an interactive video intelligence workspace.

---

Project Goal

VideoRAG is designed as an AI-powered decision engine for long-form YouTube content.

It helps users answer questions such as:

• Should I watch this video?
• Which video explains this topic better?
• What part of the video matters most?
• Where exactly was this explained?

The focus is on reducing wasted time and improving trust through source-grounded answers.

---

Features

• Multi-video comparison and cross-video reasoning
• AI-generated concise and detailed summaries
• Semantic search with timestamp-based retrieval
• Context-aware question answering from video content
• Study mode with quiz generation and learning support
• Session history with persistent workspace tracking
• Transcript generation and downloadable text output
• Clean workspace-based interface for continuous exploration

---

Tech Stack

Layer	Technology
Frontend	React, Vite, JavaScript
UI / State	React Hooks, Local Storage
Backend	FastAPI, Python
Vector Database	ChromaDB
Embeddings	Sentence Transformers (all-MiniLM-L6-v2)
LLM Inference	Groq API
Video Processing	YouTube Transcript API, yt-dlp
Deployment	Vercel (Frontend), Hugging Face Spaces (Backend)

---

Project Workflow

1. User provides one or multiple YouTube video links
2. Transcripts are extracted and segmented into semantic chunks
3. Chunks are embedded using Sentence Transformers
4. Embeddings are stored in ChromaDB for retrieval
5. Relevant context is retrieved for user queries
6. Groq LLM generates summaries, answers, comparisons, and study outputs

---

Local Setup

Prerequisites

• Python 3.10+
• Node.js + npm
• Groq API Key

---

Backend Setup

cd backend

python -m venv venv
venv\Scripts\activate

pip install -r requirements.txt

uvicorn main:app --reload

Backend runs on:

http://localhost:8000

---

Frontend Setup

cd frontend

npm install
npm run dev

Frontend runs on:

http://localhost:5173

---

Environment Variables

Create a .env file inside the backend folder:

GROQ_API_KEY=your_groq_api_key_here

---

Deployment

Frontend

Deployed on Vercel:

https://video-rag-five.vercel.app/

Backend

Deployed using Hugging Face Spaces with Docker-based FastAPI deployment.

Note: Some cloud deployments may face transcript extraction limitations due to YouTube rate limiting and bot verification on shared infrastructure.