January 9, 2026
Stars align, drama online
Show HN: Similarity = cosine(your_GitHub_stars, Karpathy) Client-side
Find your code crushes by matching Stars; skeptics say it's spooky-good
TLDR: A new browser tool turns your GitHub Stars into a personal taste profile to recommend repos and find “taste twins.” Commenters are surprisingly impressed—once-wary of “The Algorithm,” they praised spot-on picks, dropped nerd trivia, and asked for longer lists, making this a practical, fun way to discover code.
GitHub just got a spooky new mirror: [puzer]’s client-side recommender turns your Stars (those bookmarks you slap on repos) into a taste profile, then matches you with “code celebrities” like Karpathy and your own taste twins. It even draws a Skill Radar and suggests repos you’d actually want. Behind the scenes, the maker sifted roughly 1TB of public activity to map 4 million developers and trained across 300,000 projects—translation: huge data, simple results. And you can try it in your browser at this link.
The comments? Pure vibe shift. A longtime skeptic admitted, “People complain about The Algorithm but it can be useful,” after it recommended a tool they instantly loved. Another user posted a cheeky “lol” with their profile like a victory meme. A history buff dropped trivia: the same math (cosine similarity) matched people back in Usenet chat groups. The mini-drama: folks want longer “Similar to you” lists and endless scroll, and there’s a playful debate over whether Stars are true bookmarks or just flexes. Meanwhile the consensus is surprised delight: the picks feel eerily accurate, the Skill Radar is a fun brag board, and the whole thing works without shipping your data to a server. Turns out, if the algorithm actually helps, devs won’t just tolerate it—they’ll cheer.
Key Points
- •The demo builds an embedding from a user’s GitHub stars.
- •It computes similarity (e.g., to Karpathy) using cosine similarity.
- •The tool can compare and discover popular people with similar interests.
- •It generates a Skill Radar summarizing a user’s technical profile.
- •It recommends repositories and supports search for similar repositories, running client-side.