01 · MACHINE LEARNING · IOT · 3D WEB · HACKATHON (2ND PLACE)

Power Guard 360 — Predictive Maintenance for a Power Plant Feeding an AI Data Center

Overview

Built at CodeRED Orion, the University of Houston's biggest hackathon, where I led our team to 2nd place out of 70+ teams and a $500 prize for Best Pitch/Demo. Two Random Forest models watch every turbine's sensor data and flag failures with the reasons behind them. A live Arduino sensor board drives one turbine and watches for gas leaks, a 3D digital twin shows the whole plant, and the engineer's phone app takes them to the job.

Details

Key result2nd of 70+ — teams at CodeRED Orion 2026

Links

Live website YouTube demo Devpost Source code Hackathon achievement

Stack

Pythonscikit-learnSHAPFastAPIReactTypeScriptthree.jsNode.jsWebSocketsExpo / React NativeArduinoGemini APIBlender

2ndplace out of 70+ teams

96%precision on unseen data

0.90PR-AUC for failure detection

0.93macro F1 for failure type

Next projectSakina Job OS — Agentic AI Job Search & Outreach Platform →

Eight gas turbines, the AI data center, the battery yard and gas metering, with live readings and the model's diagnosis for the selected turbine.

The Random Forest flags GT-03 with a 99% overstrain risk and lists the readings that drove it. The manager decides what happens next.

Arduino UNO R4 WiFi with a motor speed sensor and an air quality sensor, posting readings to the plant server twice a second.

Our team on stage at CodeRED Orion with the $500 Best Pitch/Demo check.

The power plant and the AI data center it feeds, built for the demo.

Sense, predict, protect, see, act and fix, all in one place.