Geospatial ML
SlopeSense
Satellite and weather data fused into a landslide-risk view at 1 km² resolution, refreshed every six hours.
SlopeSense combines satellite and weather data into a probabilistic landslide-risk map at 1 km² resolution on a six-hour cycle. The pipeline uses LightGBM calibrated against NASA LHASA v2, with inputs from NASA GPM and SMAP, Sentinel-2, NOAA GFS and a Copernicus DEM. Celery and Redis handle ingestion, with Alembic, Prometheus, Docker, Nginx and GitHub Actions around it.
My resume reports 6 of 6 historical events retrospectively validated, under 120 ms p95 API latency, CAP v1.2 alerts in under 30 seconds and more than 100 pytest cases. Those are resume-reported figures, not independently rerun. I am still refining the model, alerting and reliability.