A fused, confidence-scored point-of-interest dataset combining OpenStreetMap and Overture Maps. Deduplicated, classified, and openly available across multiple countries.
About the project
AnythingPOI is an open-pipeline project that fuses point-of-interest records from OpenStreetMap and the Overture Maps Foundation into a single, clean, deduplicated dataset. Each record carries a confidence score derived from source overlap, attribute richness, and spatial uniqueness.
The project is reproducible: the full pipeline — ingestion, conflation, scoring, and tile generation — runs from open source code and public data. No proprietary services. No API keys for the core data.

OSM PBF files are downloaded from Geofabrik and parsed with PyOsmium. Overture Maps data is queried via DuckDB directly from the S3 release.

POIs are spatially indexed using H3 resolution-11 hexagons (~24 m edge). Candidates are matched on category agreement and Jaro–Winkler name similarity.

Each POI receives a confidence score (0.01–0.99). Positive signals include dual-source verification, Wikidata linkage, and contact data richness.

Output is written as PMTiles v3 (zoom 4–16) and Cloud-Optimised GeoParquet partitioned by Tier-1 category for downstream use.
Coverage
The pipeline is region-agnostic. Any area with OSM PBF coverage and an Overture Maps bounding box can be processed. Current and planned regions:
Upstream data
AnythingPOI builds on two major open geospatial datasets, with encyclopaedic verification from Wikidata.
Classification
A two-level taxonomy maps 2,116 OSM tag rules and Overture category strings into a consistent hierarchy, iteratively maintained and openly tracked.
View full taxonomy →