Reproduce “KAPSO: A Knowledge-grounded framework for Autonomous Program Synthesis and Optimization” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Alireza Nadafian, Alireza Mohammadshahi, Majid Yazdani
KAPSO treats program synthesis as an evaluator-driven loop that proposes code changes, runs them, and uses measured results to guide later attempts. It preserves experiments in Git branches, retrieves structured knowledge, and stores trace-derived lessons. On MLE-Bench, Leeroo’s Medium-category medal rate is 44.74%, versus 21.05% for R&D-Agent, while both score 68.18% in Low; this end-to-end comparison does not isolate causes.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.