SeeFix

SeeFix is a camera-first home-repair copilot for first-time fixers and distant family helpers. Point a phone at a broken fixture—our hero is a running toilet—and SeeFix reads the model plate, outlines the exact part to touch, draws the next step on your own photo, blocks unsafe steps, and verifies the fix before you call it done. Under the hood, Meta models each do real work (not a chat wrapper): • Muse Spark Open (Baseten, meta-models/muse-spark-open-1026) identifies brand/model/serial from a plate photo into typed JSON. • SAM 3.1 masks the named part (e.g. “toilet flapper”) on the user’s frame. • Muse Glimmer (open weights on Modal) is a fail-closed safety gate for gas, live mains, and scald risk. • Muse Image edits the user’s photo into an annotated step card. • Rate limits (per-IP + global daily) sit in front of paid calls so the public demo stays honest. Tagline: Camera-first repair copilot: it sees the part and checks your work. Live demo: https://seefix.vercel.app · Lab: https://seefix.vercel.app/lab · Source: https://github.com/dl88jing/seefix Team: solo — Dylan Ler, Green Robin Labs. Built during the Meta Global AI Developer Hackathon (Oct 26–Nov 4, 2026).