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

SignSight turns one phone video of a building into a complete signage audit, with redesigned signs ready to print. The problem: in hospitals, campuses and metro stations, signs are often too small to read at walking pace, low in contrast, written in only one language, or missing at the junctions where people need them most. Fixing this today means a manual audit: a designer with a clipboard spending days walking corridors. So most buildings never get audited, and visitors stay lost. How it works: you film a walkthrough of a floor. SAM 3.1 detects and tracks every sign, door plate and exit sign across the video, so each is counted once. Muse Spark scores each sign on six criteria (legibility, contrast, language access, clarity, condition, placement), watches the whole walk to flag junctions with no sign, and learns the building's own colour system. Muse Image then redesigns the failing signs in whichever languages the building needs, matching its existing style. Each sign is read from the sharpest frames of the original video; if it's still unreadable, SignSight asks for a re-capture instead of guessing, so it never invents text. Tested on the 4th floor of IIIT Delhi's R&D Building: 21 signs audited, 6 bilingual redesigns, and 4 junctions found with a missing sign, for $0.63 in API cost. A days-long audit becomes a single walk.