MARINE BUREAU · ★ HONORABLE MENTION
ZERO-SHOT VISION VERIFIES BEACH CLEANUPS
Ocean conservation, gamified — with a model that checks you actually did the cleanup
SEATTLE — Ocean pollution is a crisis, but awareness campaigns rarely change behavior for long. Static apps with manual point systems are easy to game and hard to stay engaged with. Sea Score, an honorable-mention hackathon build, answers with proof: users complete sustainability challenges, upload a photo, and a machine-learning model verifies the work before a single point is awarded.
Verification runs on Xenova OWLv2 via @xenova/transformers — a zero-shot detection model that validates photos against each challenge's keywords through text prompts alone. Because no challenge-specific training is required, any new challenge type is supported the moment it is written, gated by a confidence threshold.
HOW IT WORKS
Verified cleanups earn points and passport stamps — a visual record of cumulative impact that unlocks milestone rewards. Challenges are organized by category, from recycling to shoreline cleanup, each with its own point value.
A real-time community feed, backed by Firebase Realtime Database, shows completed challenges across all users; team leaderboards let groups compete, and a resource hub carries educational material on plastic waste and ocean ecosystems. The React Native and Expo codebase ships to three platforms from one source tree.
LESSONS FROM THE FLOOR
The zero-shot bet traded raw accuracy for flexibility — the right call for a broad challenge catalog under hackathon time pressure. Just as decisive was a singleton pattern caching the loaded model: reloading it per request would have made the app unusable. Passport stamps, the team found, built an identity around environmental action that pure point totals never achieve.