GeoAI Systems — Remote Sensing, Risk Modeling & Spatial Infrastructure
Four end-to-end GeoAI reference builds spanning satellite time-series modeling, spatial econometrics, deep-learning change detection, and high-throughput spatial data infrastructure.
A set of production-shaped reference systems demonstrating a full range of GeoAI techniques, each built as a complete pipeline — ingestion through a deployable interface, not a notebook.
Canopy — Remote-sensing crop & soil-moisture pipeline. Sentinel-2 time-series ETL against the Planetary Computer STAC API, vegetation and moisture index engineering, and an XGBoost model predicting field-level yield performance relative to crop norms.
Meridian — Spatial economic & risk analytics engine. Census, OpenStreetMap, and nonprofit-registry data fused into a multi-criteria vitality index, with a Geographically Weighted Regression model that lets economic risk vary by neighborhood instead of assuming one citywide coefficient.
Strata — Deep-learning satellite change detection. A U-Net semantic segmentation model classifying land cover from Landsat imagery, differencing two time points into a full transition matrix rather than a binary change mask.
Pulse — High-throughput spatial API. Event data pre-indexed onto Uber's H3 hexagonal grid and served through DuckDB's columnar query engine — millisecond-latency spatial aggregation over hundreds of thousands of records, with a deck.gl interface on top.
Each system includes its own architecture documentation, covering the specific engineering trade-offs behind the tool choices.