Skip to content

Hackathon note

Track, scope, and what was built for the submission.

Built for the HackCulture AI Global Hackathon, track: Build a Benchmarking Platform for AI Agents, Not Just Models. Deadline: 21 November 2026.

  • Deterministic scoring with no LLM in the score path.
  • A real Docker sandbox with an allowlisted model proxy.
  • Three adapters: fake baseline, Python loop, and Aider.
  • A public read-only leaderboard and result explorer.
  • Recorded real runs on one task and one model.

The source, README, and demo video are linked from the repository. The public deployment is read-only for real results; fake runs are self-serve.