On 1 Nov, Andrej Karpathy shared HackerTube, for which Claude filtered 43M Hacker News comments from 20 years down to 6,454 recommended videos, a site he said took about 2 hours with Opus 5.5 in Claude Code, with the data published as a public zip. A Lead SWE replicated the approach for books using Cursor, producing HackerShelf. The pipeline scanned HN comments from Jan 2024 to Aug 2026 from open-index/hacker-news, 9,437,609 comments in total; a DuckDB regex pre-filter took about 9 seconds and left 11,519 candidates; gemini-3.5-flash-lite in 288 batches of 40 found 6,076 book mentions; Open Library matched 2,212 works; and using Karpathy's rule of at least 3 people and at least 3 months, 151 books were left, with one Gemini call per book for 151 calls. The final shelf has 151 books, 723 mentions and 601 unique recommenders, sorted into 9 categories. Total time was about 2.5 hours of wall clock (7:47 to about 10:17 PM IST), with Cursor's coding portion about 40 minutes. Tokens were 1,968,177 in and 623,943 out, costing $0 on Gemini's free tier.