Harvard physics professor Matthew D. Schwartz released BootLoops 1.0, an open-source toolkit designed to connect large language models with scientific research tasks. The software acts as a modular interface between researchers and AI models like Anthropic’s Claude, Google’s Gemini, or OpenAI’s ChatGPT.
The toolkit focuses on mathematical physics, specifically the computation of integrals. In a three-month testing period, the team evaluated 400 candidate problems, resulting in 36 scientific manuscripts co-authored by 19 researchers. The system successfully computed 30 integrals, including 15 previously unknown elliptic Feynman integrals.
Beyond particle physics, the toolkit is applicable to 18–22 scientific fields, such as cosmology, ecology, and population genetics. The system generates Python scripts to ensure results can be verified with arbitrary precision. Development began in December 2025 using Claude Opus 4.5.
Schwartz emphasizes that BootLoops is intended strictly as research instrumentation. The toolkit is not designed for use in safety-critical, clinical, regulatory, or actuarial decision-making processes.
