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

SiliconLoop turns a photo of electronic parts and one plain-English sentence into a working Arduino gadget that is proven before you touch a wire. Muse Spark Open does every AI step: it recognises the components in the photo, plans the gadget as machine-checkable behaviour rules, writes the firmware, and answers questions about the build. Anything electrical is decided by deterministic code, never by the model: pin assignment, resistor values, the power budget and an electrical safety check that refuses unsafe designs. The compiled firmware then runs in a simulated Arduino (avr8js) against every behaviour rule. When a rule fails, Muse Spark Open receives the failure report, repairs its own code, and the test runs again. Only firmware that passes every rule is released. The user gets three things that always agree: a live digital twin running the exact verified binary in the browser, the firmware with one-click flashing to a real board over Web Serial, and a step-by-step build report with wiring, pin table and safety notes. In a 27-wish evaluation, SiliconLoop produced verified firmware for 22 of 22 buildable gadgets, against 6 of 22 for Muse Spark Open on its own. It refused 5 of 5 unsafe designs, at about 4-6 US cents and 30-45 seconds per build.