Anthropic x BAINSA Hackathon - Mathematical Autoresearch
Second place by number of problems solved. Our team built AI agents to explore mathematical conjectures and develop proofs, solving 20 of 24 problems. The remaining four were open mathematical problems.
Research approach
The agents searched for counterexamples and developed proof arguments. We worked on angles between lines, paths on hypercubes, Bulgarian solitaire, and disjoint congruence classes.
Proofs and checks
We used Lean and exact Python computations to check the work. The repository includes written proofs, verification scripts, and computational certificates, with complete proofs distinguished from finite checks and unfinished arguments.