SharkWatch
Built at Meta Global AI Developer Hackathon · Oct 26, 2026 · Remote
Millions of people go to the beach every year to swim, surf, and enjoy the water, but many are not comfortable getting in because of sharks. Not because attacks happen every day they don't. It's because you cannot see what's under the surface, and nobody can warn you in time. A shark is almost invisible from shore and a person-to-person warning chain is too slow by the time the message reaches the tower, the moment is gone. So people stay near the shore, or stay out of the water altogether. The solution SharkWatch SharkWatch watches drone footage from above, understands what's in it, and sends a confirmed alert to authorities in seconds, with an evidence photo and a risk level. SharkWatch analyzes aerial footage in three steps: Seen: Meta SAM 3.1 traces sharks and people in six-second video chunks. A track must appear in eight frames before it counts, which filters out flashes of glare. Named: Muse Spark checks each trace: is it a shark, or a ray, dolphin or kelp? Acted on: A risk engine scores each shark on size, species and distance to people and shore, and a species guess never lowers the risk. Lifeguards get one an alert with an evidence photo.