HackerTube

HackerTube: An AI-Built Video Library from 19 Years of Hacker News

HackerTube is a remarkable demonstration of what an AI coding agent can build end to end: a full video platform that surfaces the YouTube talks, lectures, and documentaries that Hacker News commenters have recommended to each other since 2007. The entire pipeline behind HackerTube was constructed by Claude working in Claude Code — data collection, comment labeling, video cards, semantic search, recommendations, and the HackerTube site itself.

The results speak for themselves. HackerTube processed the complete public Hacker News archive, extracted over one hundred thousand YouTube links, and kept the 6,454 videos that real people recommended repeatedly across different years. An AI model then read each of the 39,834 linking comments, judged whether it was a genuine recommendation, and wrote a card for every video explaining why HN users love it.

What makes HackerTube special as a product is the experience: a daily mix of HN classics, hidden gems, and recent favorites; semantic search that understands topics rather than exact keywords; playlists, likes, and history that all work without an account. Every page on HackerTube shows the real Hacker News conversation behind each pick. The whole dataset is also open under ODC-By 1.0, so developers can fork it, remix it, or hand the zip to their own AI agent to build something new. HackerTube is free to use and a masterclass in shipping a complete, useful product with AI assistance.