HakkTaxi: ride-share demand prediction (Microsoft Azure APAC Hackathon)

Jakarta's ride-hailing market suffers from supply-demand mismatch that manifests as surge pricing during predictable high-demand windows (morning commute, Friday evening, rain events). Driver repositioning is reactive -- drivers respond to surge pricing signals that are already 5-10 minutes late. A demand heatmap that forecasts surge zones 15-30 minutes ahead would allow proactive repositioning, reducing surge duration and improving driver earnings predictability.

A 48-hour competition window eliminates access to the proprietary trip telemetry that production demand forecasting requires. The core feature engineering challenge: constructing a credible demand signal from proxy variables -- historical aggregates, weather data, and static POI density -- that generalises across Jakarta's heterogeneous urban geography. CBD office clusters, university campuses, market districts, and transit hubs each exhibit fundamentally different demand dynamics that a single model must capture without overfitting. The additional constraint: a system demonstrable in 5 minutes to non-technical judges while withstanding commercial scrutiny from Microsoft product leadership on the jury panel.

Within 48 hours: designed a demand heatmap and surge prediction system using gradient boosting (XGBoost) on historical trip logs, weather data, and Points of Interest (POI) density features per geospatial hex cell (H3 grid). Temporal features: hour-of-day, day-of-week, public holiday flag, historical demand rolling averages. Geospatial features: POI category counts (office density, transit hub proximity, entertainment zones). Integrated with Azure Maps API for live visualization as a real-time demand heatmap overlay.

Competed against teams across Southeast Asia and Australia. Won Regional Champion across APAC -- validated against international jury with Microsoft engineering and product leadership as judges.