Skip to Main Content

Whiteboard2Repo

Built at Meta Global AI Developer Hackathon · Oct 26, 2026 · Remote

Whiteboard2Repo — Demo video

Every system starts as boxes and arrows on a whiteboard. Then the photo rots in a chat thread, someone re-types the design into boilerplate, and the rules scribbled in the corner ("no double booking!") get lost. Whiteboard2Repo turns that photo into a running, tested repository and keeps it in sync when the board changes. Muse Spark Open reads the photo into a typed architecture graph (components, labelled arrows, and each handwritten note as a testable invariant), and you can fix misreads before building. A deterministic scaffolder emits a green FastAPI + docker-compose skeleton, then a Muse Spark Open tool-calling agent implements every endpoint and writes tests until pytest passes. For each rule, the model acts as a mutation-testing engine: it breaks the rule in a throwaway copy and the tests must fail (mutant killed). Re-snap: photograph the updated board; Whiteboard2Repo diffs it, changes only what moved, keeps existing tests green, proves each new rule, and opens a GitHub pull request whose description is the verification report. In the run shown in the video: 16 tests passing and the "no double booking" mutant killed; after re-snap (+Payments API, +"max 3 active bookings per user") 36 tests passing. Real PR: https://github.com/taku629/whiteboard2repo-demo/pull/1. Try it: https://whiteboard2repo.onrender.com. The sample boards are AI-generated test images.

Team