Radesh Govind

HyperLearn: verifiable adaptive learning, built from your own material. Students learn from slides, papers, lecture recordings and past exams, but generic AI tutors ignore that material and can't be trusted to stay faithful to it. HyperLearn builds a course around your sources: every explanation is cited down to a highlighted page region or an audio timestamp. When something doesn't click, change it. Select a paragraph and ask "simplify this and give me an interactive example". Agents generate, verify and repair: schema-validated patch operations over typed blocks (surrounding content stays intact), model-based source checks on every new sentence, deterministic math checks on widget claims, browser execution tests and screenshot review. Failures feed a bounded repair loop, and unsupported parts are removed. Content-addressed versions keep provenance and make every change reversible. Practice in the style of your past exam (25 question types, rubric grading). A mistake triggers concept-level diagnosis and a proposed learning-path change that you choose to accept. Share your course: others can follow your updates or make their own copy. Eval: 93.2% of rewritten sentences were supported by their cited source (82/88 across 18 real requests). Results and known limitations are public in the repo. Verification reduces errors; it doesn't guarantee correctness. Live: https://hyperlearn.radesh-govind.com (demo mode, no API key needed). Built during the Meta Global AI Developer Hackathon (see HACKATHON_LOG.md).