First72(Solo)
Built at Meta Global AI Developer Hackathon · Oct 26, 2026 · Remote
First72 turns before and after flood imagery into a ranked, uncertainty-aware dispatch plan in seconds. It maps what is newly flooded, ranks zones by people affected with ranges instead of single numbers, plans flood-aware routes, and flags zones reachable only by boat or air. An audit agent cross-checks the zones it is least sure about with independent evidence and flags unresolved ones “verify on ground”. A verifier rejects any model answer that cites a zone that doesn’t exist. It writes a field brief in English and Bangla that carries the uncertainty through. Built solo during the hackathon, with Claude as a coding assistant. In my synthetic benchmark, flagged zones are the wrong ones: zone IoU 0.46 against 0.83 for confirmed zones.