Tileserver Polars (Rust Optimized)

Concept

A dynamic vector tile server for massive geospatial datasets. It serves Mapbox Vector Tiles (MVT) from millions of points with sub-second latency, enabling interactive visualisation in Kepler.gl and MapLibre without pre-rendering a static tile set. Python (FastAPI) handles the HTTP layer; Rust (via PyO3) does coordinate projection and Protobuf encoding; Polars holds the data and runs the spatial filters.

The argument was that pre-generating tiles for a large dataset is both storage-intensive and inflexible: every change of filter or zoom strategy means re-generating the pyramid. Dynamic tiling backed by columnar in-memory filtering trades a small per-request cost for the ability to change the query at any moment.

Architecture

• API layer — FastAPI serving MVT/PBF tile requests over HTTP.
• Compute layer — A Rust extension via PyO3 for per-point coordinate projection and Protobuf MVT encoding, where Python overhead would dominate.
• Data engine — Polars for vectorised spatial filtering and aggregation against the in-memory dataset.
• Consumers — Mapbox Vector Tiles, consumed by Kepler.gl or MapLibre.

Quick Facts