// side project
FireSight
Wildfire risk intelligence, built for the people who need it most.
A geospatial risk platform that combines live weather, drought history, and Sentinel-2 satellite imagery into a reproducible 0–100 Fire Danger Index for any location, with historical replay and an ML evaluation pipeline for improving the underlying risk model.
Live Data Fusion
FireSight brings together hourly weather, long-term climate data, drought state, Sentinel-2 vegetation imagery, and NASA FIRMS thermal detections into a unified location-level feature pipeline. Satellite pixels are read directly from cloud-hosted imagery rather than downloading entire scenes.
Risk Engine
A deterministic scientific engine calculates weather danger, drought severity, vegetation moisture, and other environmental signals before combining them into a transparent 0–100 Fire Danger Index. Missing or invalid data is explicitly surfaced rather than silently imputed.
Time-Aware Analytics
Every assessment preserves the inputs available at that point in time, enabling historical reconstruction without accidental look-ahead. FireSight can distinguish between today's live assessment, retrospective analysis using improved historical data, and operational replay using only information that would have been available at the original decision time.
Production & ML Platform
The system wraps the risk engine in a production architecture with PostgreSQL/PostGIS persistence, background jobs, APIs, monitoring, migrations, containerized deployment, and automated testing. A separate ML pipeline builds datasets, trains candidate models, evaluates them against the deterministic baseline, and supports controlled model promotion and rollback.
Live demo
Try FireSight yourself — here's the live dashboard
The production system is running. Explore real-time wildfire risk scores, data fusion, and the ML pipeline outputs directly in the app.
Open FireSight dashboard →