✕
GeoQL · Planet
An open, planet-scale geospatial datalake — built from raw public data to a globe you can fly. Every number below is real.
53,000,000
global Overture places, named & served static
2.4 billion
building footprints, tiled on-demand per viewport
1,359,562
FEMA Special Flood Hazard Area polygons
51
US states + DC — national flood coverage
The lakehouse
- Medallion: raw → bronze → silver → gold → tiles
- Storage: S3 + Apache Iceberg + GeoParquet, queried with DuckDB
- Conflation (the moat): spatial point-in-polygon joins — "what governs / floods / applies at this lat,lng"
Tiling — map-reduce, not a cluster
- National flood tiled as 469 grid-shard cells in parallel across 32 cores
- Turned a 7.5-hour serial hang into a ~30-minute parallel run
- Per-cell PMTiles tile-join'd into one national pyramid
Serving — $0 egress
- PMTiles on Cloudflare R2, HTTP range-read from the edge (~4.7 GB of tiles)
- The browser fetches only the handful of tiles in your viewport
- Tiles at tiles.geoql.dev; this page on Cloudflare Pages
Compute — next to the data
- Ephemeral in-region EC2 workers, self-terminating; cloud-to-cloud, never via a laptop
- Reads Overture + FEMA NFHL public data anonymously
- Stack: DuckDB · tippecanoe · GDAL · MapLibre GL
Open data in → fly-the-planet out. The point-query API (api.geoql.dev) is next.