Equipment Inspection System for Saudi Aramco

Key Metrics

Problem

Industrial sites lose money when equipment fails without warning. Inspection teams had historical records, but no fast way to turn them into a prioritized risk view, so high-risk assets looked the same as healthy ones until something broke.

My Role

Full-stack contributor on a 5-person Agile team under a 48-hour clock. I built the React Router dashboard (risk heatmaps, asset views, analytics) and integrated the FastAPI prediction service so inspectors could go from data → risk → action in one UI.

Approach

We engineered features from 1000+ inspection records and trained a Random Forest classifier (92% accuracy), explainable enough to trust in a hackathon demo, fast enough to retrain. Predictions ship through FastAPI; the UI maps scores onto heatmaps and alerts. A hybrid fallback keeps the dashboard usable if the model service drops, because a blank screen is worse than a degraded prediction path.

Tradeoffs

Classical ML beat a heavier neural net for this timeframe: we could train, debug, and explain it in hours. The fallback path sacrifices some prediction richness for uptime, the right call when judges are clicking live. Scope was ruthlessly cut to heatmaps + assets + alerts rather than a full CMMS.

Outcomes

• 92% accuracy on 1000+ inspection records
• Inspectors see risk spatially via interactive heatmaps
• Production-shaped demo shipped in 48 hours with 100% prediction-path uptime design

Tech Stack