SEA SCORE
Ocean conservation, gamified — with ML that verifies you actually did the cleanup.
An eco-conscious mobile app that rewards users for taking real environmental action. Users complete sustainability challenges, upload photo proof verified by a zero-shot ML model, earn points, and engage with a community of eco-active people.
The Challenge
Ocean pollution is a critical crisis, but most awareness campaigns fail to drive sustained behavioral change. Users need immediate, verifiable feedback on their real-world actions — and a reason to keep coming back. Static apps with manual point systems are easy to game and hard to maintain engagement with.
The Solution
Sea Score uses a zero-shot ML model to automatically verify photo submissions — no challenge-specific retraining needed. Users earn points for verified cleanups, collect passport stamps, redeem rewards, and follow each other's progress in a real-time community feed, making sustainability fun, social, and rewarding.
Zero-Shot ML Verification
The backend uses Xenova OWLv2 (owlv2-base-patch16) via @xenova/transformers to validate photo submissions against challenge keywords. Because it's zero-shot, any new challenge type is supported immediately — no model retraining required.
Challenge & Passport System
Each challenge has a description, category (recycling, cleanup, etc.), and point value. Completing challenges awards stamps to the user's Passport, giving a visual record of cumulative environmental impact and unlocking milestone rewards.
Community & Leaderboards
A real-time community feed shows completed challenges across all users. Team leaderboards let groups compete on scoreboards. A Resource Hub provides educational content on reducing plastic waste and protecting ocean ecosystems.
Frontend
React Native · Expo · Figma UI
Backend
Node.js · Express · multer
AI & ML
Xenova OWLv2 · @xenova/transformers · Zero-Shot Detection
Auth & DevOps
Firebase Auth · Firebase Realtime DB · Docker + Compose
Zero-Shot vs Fine-Tuned Models
Using a zero-shot model like OWLv2 traded raw accuracy for flexibility — any new challenge category works without retraining. For a hackathon with a broad scope of challenges, this was the right trade-off.
Firebase for Real-Time UX
Firebase Realtime Database made the community feed and leaderboard trivially easy to implement with live updates, removing significant backend complexity under time pressure.
Gamification Psychology
Passport stamps and visible progress created a sense of identity around environmental action. The social leaderboard added competitive motivation that pure point systems lack.
ML Model Caching
The singleton pattern for model loading was critical — loading the model on every request would have made the app unusable. Pre-downloading and caching kept inference fast.