Getting ArXiv Sanity Papers via RSS (2022) (den.dev)

arxiv-sanity-feeds is an open-source utility that turns Andrej Karpathy’s arxiv-sanity-lite views into consumable RSS feeds. The author implemented a GitHub Actions job that runs daily, scrapes the structured JSON blob embedded in arxiv-sanity-lite (so no brittle HTML parsing), and runs a Python script to reformat entries (authors, id, title, summary, tags, time, weight, thumb_url) into RSS. The generated feeds (home, most-recent-week, random-last-week, etc.) are pushed to a DigitalOcean Spaces bucket via boto3 (S3-compatible API) and served through a CDN and custom domain. The pipeline also runs an automated W3C RSS validation step wrapped in a Bash Action that checks the validator’s response and fails builds if feeds become invalid. This matters for the AI/ML community because it restores the simple, ubiquitous workflow of following curated arXiv discovery pages in any RSS reader (Feedly, NetNewsWire, etc.), enabling researchers to integrate paper discovery into existing reading and automation pipelines. Technically notable choices: leveraging the JSON view exposed by arxiv-sanity-lite to avoid fragile HTML scraping, using GitHub Actions for scheduled ETL, and publishing to S3-compatible storage for global consumption. The project is on GitHub and invites contributions; a planned future enhancement is a serverless query layer to generate filtered feeds (e.g., only ML papers).