Marine-AI

Fiberglass passenger boats are the backbone of inter-island transport in the Philippines, and almost none of them are instrumented. Fuel burn, engine health, and route choice are decided by feel. The efficient answer is usually a new vessel, which is not an answer at all for an operator running a twenty-year-old hull.

So the constraint was retrofit-only: whatever we built had to bolt onto boats that already exist.

Three sensor systems feed three AI modules running in parallel, each answering a different question, and all three converge on one bridge display so the skipper reads a single screen rather than three dashboards.

Speed Optimization uses XGBoost exported to ONNX for optimal throttle settings. Route Optimization uses gradient-boosted models over live track data. Predictive Maintenance uses a PCA autoencoder to flag anomalies before they become failures.

Speed Optimization runs XGBoost exported to ONNX, producing optimal throttle settings. Route Optimization uses gradient-boosted models over live track data. Predictive Maintenance uses a PCA autoencoder, framed as anomaly detection rather than classification, since a retrofit fleet has no labelled failure history to train against.

All three converge on one bridge display: live route tracks, throttle guidance, and an auditable CO₂ emissions layer, so the skipper reads a single screen instead of three dashboards.

2nd Place out of 24 teams in the National Open Professional Category at the National AI Hackathon 2026, competing as AI Engineer and Full Stack developer for Team SOLMATE. The system also produces an auditable CO₂ emissions layer, so efficiency gains can be reported rather than merely claimed.