Melbourne CBD Digital Twin
A real-time 3D digital twin of Melbourne CBD built on CesiumJS, deck.gl, and pure public data. No server runs between data fetches — GitHub Actions cron jobs ingest GTFS-RT tram positions, pedestrian sensor counts, BOM weather, and EPA air quality, writing flat JSON to Cloudflare R2 every five minutes. The browser fetches, diffs, and renders only what changed.
Stack
- 3D renderer: CesiumJS with OSM Buildings and world terrain
- Data overlay: deck.gl IconLayer (trams) + HeatmapLayer (pedestrian density)
- Storage: Cloudflare R2 — zero egress fees
- Frontend: React 19 + Vite, served via Cloudflare Pages
- Ingestion: GitHub Actions cron, ~1,728 min/month of free tier
Cost model
| Service | Free tier | Usage |
|---|---|---|
| Cloudflare R2 | 10 GB storage, zero egress | ~5 GB tiles, ~3 TB/month transfer |
| Cloudflare Pages | Unlimited bandwidth | Full site delivery |
| GitHub Actions | 2,000 min/month | ~1,728 min/month |
| Domain | — | ~$1/month |
Total: ~$1/month. AWS S3 at equivalent usage would cost ~$255/month in egress alone.
Architecture
Three independent layers: static geospatial assets (processed once, cached aggressively), live feed ingestion (GitHub Actions → R2), and a browser-side renderer that polls R2 every 30 seconds and updates only what changed. DuckDB-WASM handles analytical queries entirely in the browser with no backend required.