How I Built an AI-Powered Hacker News Book Recommendation Site in 2.5 Hours
Karpathy built HackerTube in about 2 hours with Claude Code. I rebuilt it for books with Cursor. Here's my honest time, cost, and everything that broke.
On November 1st, Andrej Karpathy released HackerTube (https://hackertube.org/) which filtered 43 million Hacker News comments from the past 20 years down to 6,454 book recommendations. He accomplished this in about 2 hours using Claude Code with Opus 5.5. Karpathy's data is available as a public zip file. Intrigued by the possibility of applying the same technique to books, I decided to use books instead of videos and Cursor, a tool I frequently use for writing, as the platform.
The resulting site is HackerShelf (https://ishankdev.github.io/hackershelf/) with code available on GitHub (https://github.com/IshankDev/hackershelf).
Karpathy's process involves scanning all Hacker News comments from 2007 to 2026, keeping only those containing YouTube video links, and filtering for mentions made by at least three different people in at least three different months. Claude Sonnet 5.5 labels each mention with a stance, topics, a reason and a quote. The extracted information is then organized into one to three categories.
After some manual steps, including obtaining a Google Cloud API key for YouTube metadata and setting up a VPS and domain, the final outcome is a website displaying the curated list.
My version of the site follows the same concept but focuses on books. The source data includes Hacker News comments from January 2024 to August 2026, totaling 9,437,609 comments. Initial filtering using DuckDB regex for book links and phrases like "must read," "worth a read," and "book called" took about 9 seconds and left 11,519 candidates.
Using Gemma-3.5-flash-lite, 6,076 book mentions were extracted in 288 batches of 40. Open Library matched 2,212 works. Applying Karpathy's rule of at least three people mentioning a book in three different months, 151 books remained. Cards were generated for each book using Gemini, resulting in a site with 151 books, 723 mentions and 601 unique recommenders, organized into 9 categories.
In total, the project took approximately 2.5 hours to complete, including manual steps such as creating the repository, granting Cursor GitHub access, and setting up GitHub Pages.
Written by urgent.news from HackerNoon's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
