Skip to Main Content

Lusot

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

Lusot — Demo video

Every rainy season, Metro Manila asks: how deep is it at this spot? Lusot answers in MMDA's gauge words (gutter, knee, tire, waist…). A photo from a dry spot becomes a level and, where the level allows, MMDA's passability class, with outlines showing which wheel or person set it. Readings become live map pins, and Ask Lusot answers typed or spoken questions from nearby reports only. When a photo can't support a reading, Lusot says "can't tell". Code, not the models, sets the level, by rules written before any result was seen. A wheel-geometry method didn't pass its pre-set release gate, so it stays off. Results (indicative, small samples): 0 of 15 flood-free street photos reported as flooded; 8 of 9 levels matched labels on open-licence photos; 1 of 6 sealed photos got a level, read deeper than its label. Not affiliated with MMDA. It is not advice to cross.

Team