Arxiv Sanity Lite

Arxiv Sanity Lite is a lightweight rewrite of Karpathy’s original arXiv discovery tool for tracking machine-learning research. It periodically polls the arXiv API for new papers and stores them locally for browsing. Users can tag papers they find relevant and receive recommendations tailored to each tag. Recommendations are generated with support vector machines trained on TF-IDF features extracted from paper abstracts. The web interface supports searching, ranking, sorting, and filtering the indexed collection. Optional daily emails can surface newly added papers that match a user’s interests. The project can be self-hosted with a Flask server and scheduled background update jobs.