Skip to Main Content

RoadScout

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

RoadScout — Demo video

RoadScout turns a phone on your dashboard into a road-hazard reporter. You upload a drive (video plus a GPS track). SAM 3.1 finds hazard candidates (debris in the lane, damaged guardrails and signs, faded lane markings, potholes). Muse Spark checks each candidate and writes a severity and a plain-language description. Deterministic code then geotags each hazard, merges repeat sightings and checks later drives to see whether a hazard is still there. Faces and license plates are blurred before any image reaches the screen. A person reviews each hazard and gets a prefilled VDOT (Virginia) or MDOT SHA (Maryland) report to submit on the agency's own site. RoadScout never submits anything itself. On two real drives (49 minutes of video) it found 5 reports of 4 distinct hazards, all confirmed by hand (debris on the shoulder, such as a tipped construction barrel), for about $14 of API spend. It missed the one known pothole, a small one among heavy cracking, so we do not claim pothole detection, and recall is not independently measured. Live read-only demo: https://roadscout-demo.onrender.com/

Team