About this project

air.grid is a one-day challenge, started and finished on June 1, 2026. Built across three Claude Code sessions — each running until usage limits reset — the goal was simple: prove that a developer with the right AI tooling can go from an empty repo to a deployed, data-rich atlas in a single day.

The project wasn't meant to be ambitious. It was meant to be fast. Schema design, data ingestion, geospatial proximity joins, interactive frontend, QA, and deployment were all orchestrated through Claude Code's multi-agent system. No shortcuts on the data: 268,980 EPA-reported facilities, 15,897 live air quality sensors, 108,336 schools, real geospatial joins computed with scipy's cKDTree — not mocked, not estimated.

The Grid Analysis page surfaces what those joins reveal: which facilities emit the most, which regions carry the heaviest load, which schools sit closest to emitters, and which majority-minority Census tracts overlap with the highest-emitting sites. All of it is public record. This just puts it in one place.