AeroAnalytics.

// Context

The WHO links ambient and household air pollution to roughly 7 million premature deaths a year. AeroAnalytics gives the public an at-a-glance view of AQI, key pollutants, forecasts, and hazard alerts, turning heterogeneous scientific data into something a non-expert can read in seconds.

What we built .

The team web-scraped NASA TEMPO data, built an ETL that harmonizes disparate sources spatially and temporally, trained a short-term forecasting model, and shipped a working demo for the judges, end to end in 48 hours.

The frontend (my part) .

The flagship view is a continent-scale 3D pollution heatmap: a deck.gl HexagonLayer with GPU-accelerated aggregation over a MapLibre dark-matter basemap, with three zoom-aware interpolations (radius, elevation, coverage) so it stays legible from continent to city scale. Clicking the map deep-links to a per-location dashboard at /dashboard/:lat/:lng. The URL is the source of truth, so views are bookmarkable and shareable.

State runs on Redux Toolkit + RTK Query (30s freshness, reconnect refetch, friendly error transforms). ECharts time-series charts overlay a dashed amber forecast line that splits at the current hour, so users see the boundary between observed and predicted at a glance.

What I did.

• deck.gl heatmap integration: HexagonLayer, GPU aggregation, custom lighting, zoom-aware radius/elevation/coverage interpolation, selected-point marker.
• CSV/API pollution data loader with per-source client-side caching.
• Dashboard wiring: URL params → Redux → RTK Query → ECharts, with a forecast/historical mode state machine.
• Sidebar, filter panel, location detail, quick-jump cities, theme toggle, responsive collapse.
• API client with RTK Query and a custom transformErrorResponse for friendlier empty states.

Notable decisions.