YouTube RAG Scraper turns channels into RAG datasets

The app is the demo; the data pipeline is the product. For niche RAG, the hard part is not the model, it’s turning noisy creator video into something stable enough to retrieve.

• –The repo covers the annoying middle layer most demos skip: channel, playlist, and video scraping, transcript extraction, cleanup, chunking, rate-limit handling, resume support, and export.
• –Sentence-aware chunking plus FAISS makes it immediately useful as an offline knowledge base or semantic search index.
• –A CLI is the right shape here because it stays composable for teams that already have their own embedding stack or vector DB.
• –Coffee education content is a strong source for this kind of assistant: it’s dense, expert-driven, and much richer than most written guides.
• –The post is also a reminder that useful AI products often start as internal data plumbing before they become a standalone app.