What might a video catalogue look like if it began with years of recommendations from a community? AI researcher Andrej Karpathy’s HackerTube collects videos recommended in Hacker News discussions and presents them in a browsable interface.
From comments to a catalogue
In his October 31, 2026 account, Karpathy says the project downloaded and filtered Hacker News comments and processed the remainder using a Claude Code workflow. He reports finding around 40,000 recommendation comments and 6,454 videos across 150 topics, spanning 2007 to 2026. Those figures describe his account of the project, not an independent guarantee of each recommendation’s quality.
The site offers categories and keyboard shortcuts. Karpathy says personal preferences are kept in the browser session without requiring an account, and that the displayed thumbnail can be a frame from the middle of a video rather than the uploader’s promotional image.
What did the agent do?
Karpathy says he used the Claude Code desktop app with Opus 5.5 at medium effort, then ran Sonnet in a workflow over the filtered comments. He also lists three manual steps: creating a Google Cloud API key for video metadata, starting a server and buying the domain. His approximate two-hour build time is self-reported.
The project also makes its underlying data available for export, giving others a starting point for their own interface. A catalogue built from community recommendations offers a different discovery path; viewers still need to consider whether an individual video remains accurate and current.
