Built a browser application for practicing questions from AI benchmarks. Users can filter by subject and answer format, review reference answers, and resume practice without an account.
Multiple-choice answers use published answer keys. Numerical math answers accept equivalent values. Written and code responses use self-assessment, kept separate from automatic scores. Each benchmark links to its source and explains its scoring.
I built the application and its browser-based progress storage. The live collection contains 10 benchmarks and 45,056 questions.