YT Navigator is an AI-powered application designed for researchers, students, and content creators to efficiently explore and extract insights from YouTube channels. It allows users to semantically search and chat with a channel's video content, providing precise timestamps for relevant segments, thereby saving significant time compared to manual review.

How It Works

The application first retrieves and stores video metadata in PostgreSQL and processes video transcripts by splitting them into segments, generating vector embeddings, and storing them in PGVector. This dual-database approach enables both keyword and semantic search. Queries are processed using a combination of semantic search (Sentence Transformers) and BM25 keyword search, with results reranked by a cross-encoder model for relevance. The chat functionality leverages the ReAct framework with models like qwen-qwq-32b and llama-3.1-8b-instant from Groq, using tools to query the vector database and mitigate hallucinations.

Quick Start & Requirements

• Installation: Clone the repository, create a virtual environment, install dependencies (pip install -e .), set up a PostgreSQL database, copy .env.example to .env, run migrations (python manage.py migrate), and start the server (make dev or make prod). Docker installation is also supported (make build-docker, make run-docker).
• Prerequisites: PostgreSQL database, Python 3.x.
• Resources: Scanning up to 100 videos per channel.
• Links: GitHub Repository

Highlighted Details

• AI-powered semantic search and natural language chat with YouTube channel content.
• Combines PostgreSQL for metadata and PGVector for transcript embeddings.
• Utilizes LangGraph, Sentence Transformers, BM25, and Groq models (qwen-qwq-32b, llama-3.1-8b-instant).
• Provides precise timestamps and transcript segments for relevant content.

Maintenance & Community

The project is authored by wassim249. Contributions are welcome via pull requests. A roadmap is available, indicating planned features like Docker support and testing.

Licensing & Compatibility

Licensed under the MIT License, permitting commercial use and integration with closed-source projects.

Limitations & Caveats

The current version supports scanning up to 100 videos per channel. The roadmap indicates that playlist and YouTube Shorts scanning are not yet supported.