Inspiration
Most of us didn't grow up in London. We arrive at 18, live inside a bubble of campus, Richmond Row and one bus route, and four years later the vast majority of us leave. We spend four years in a city and never actually meet it.
We already explore cities obsessively, just in games. Open-world games make you want to see what's behind the fog. So I asked: what if London was the open world?
What it does
Mustang City is a GTA 6-inspired mobile web app that turns London, Ontario into Forest State, an open-world map covered in black fog. The fog only clears where you've physically walked. Western's campus is your home base, Mustang City, and everything else is waiting to be uncovered.
- Fog of war: your real GPS path clears the map, cell by cell, with smoothed, feathered edges.
- 3D Mustang City: all 160 campus buildings are modelled in 3D, with day and night versions. Tap any building for its real history.
- Photo missions: The Goose, an original crime-boss quest-giver, sends you to real places with real history, like the house where insulin was conceived. Snap a photo to prove you went, and Gemini checks it. Then: MISSION PASSED.
- Rep, ranks and neighbourhoods: earn Rep, unlock badges, and watch "Entering Old East Village" banners as the city opens up.
- Live LTC buses: real-time buses on the map, plus "Route 2 in 4 min" right on your mission card. The app detects bus rides too.
- Social: compare fog with friends, fight for neighbourhoods by faculty, and pull three-player heists that get strangers to meet.
- Forest FM: an in-game radio host that reads facts-only scripts in a Gemini TTS voice.
- Geo-pinned tags: leave notes on places for whoever comes after you.
- Class Run: your daily walk to class earns Rep for being on time. Your schedule never leaves your phone.
- Graduation Wrapped: a story-style recap of the fog you cleared, in the order you cleared it, with the places and the people. Not just a degree, but a map of your four years in London.
Graduate knowing your city.
How I built it
- Frontend: Next.js (App Router), TypeScript, Tailwind v4 and Motion, shipped as a PWA for the iPhone home screen. On a laptop it renders inside a phone frame.
- Map: one Mapbox GL JS instance on the Standard style. The fog is an H3 hex grid merged into a single polygon and smoothed with Chaikin's algorithm.
- 3D campus: I wrote a script that turns building footprints into glTF models, with separate day and night files because Mapbox ignores emissive strength on material glow.
- AI: Gemini handles photo verification, mission briefings in The Goose's voice, and radio TTS. A strict rule covers all of it: facts come from stored, sourced text. Gemini may restyle a sentence but never add a fact, and a guard checks every number.
- Transit: LTC's static GTFS is preprocessed into a compact dataset, and the live bus positions and arrivals come from GTFS-Realtime.
- Auth and data: Auth0 with a post-login Action that only allows @uwo.ca accounts. Data lives in Tiger Data (Postgres + TimescaleDB): location pings go into a hypertable with a continuous aggregate for leaderboards.
- Hosting: Vercel. Weather comes from Open-Meteo.
- Process: I built this solo with AI coding agents, writing a spec for each of 20 features before building it.
Challenges
- Getting custom 3D models into Mapbox: I found out the hard way that vertex buffers can't be interleaved, indices must be uint16, and model URLs must be absolute. Clip layers also hid my own models until I scoped them to the basemap.
- Buildings split at tile edges: Mapbox vector tiles cut buildings at tile boundaries, so matching footprints to my models meant testing vertex overlap both ways.
- Fog that looks good: raw hexagons look like a spreadsheet. Merging, smoothing and feathering them took several passes.
- Live transit from the cloud: the LTC realtime server reset TLS connections from the build machine, so I added an HTTP fallback.
- Accuracy vs. personality: I wanted The Goose to be funny without making up history. That's why the fact guard exists.
- Privacy: there's no live location sharing. Friends only see where you've been (cleared fog), never when, and never where you are right now.
What I learned
- How to bend a map engine into a game engine: fog, a HUD, mission flow and a cinematic intro camera.
- How to generate 3D glTF from scratch, and how to work with GTFS and GTFS-Realtime.
- How to design around time-series data with TimescaleDB hypertables.
- How to keep an LLM useful without letting it make things up.
- How to scope ruthlessly as a solo builder: one strong image (fog clearing as you walk) beats ten half-finished features.
What's next
Phone photogrammetry for photo-real campus facades, student-written missions at scale, and local businesses sponsoring missions to get students off campus and into the city.
Built With
- auth0
- claude-code
- gemini
- gltf
- google-gemini
- gtfs
- gtfs-realtime
- h3
- javascript
- mapbox
- mapbox-gl-js
- motion
- next.js
- node.js
- open-meteo
- postgresql
- pwa
- react
- speckit
- tailwind-css
- tiger-data
- timescaledb
- typescript
- vercel
- web-speech-api