KerbQuest

Why I built it

Brisbane City Council runs scheduled kerbside "clean-up" collections suburb by suburb, a few times a year. Locals, verge hunters, drive around during collection weeks looking for discarded furniture, tools and other reusable stuff before it gets picked up as landfill. There was no good way to know which suburbs were actually live this week, which streets you'd already covered, or where you'd spotted something worth going back for, so people relied on memory, guesswork, and re-driving the same roads. I'd been doing exactly that myself, so I built the tool I wanted.

What it does

• Live clearout map — this week's and next week's collecting suburbs glow on the map, pulled from Council's public schedule. A Fog of War mode dims everything that isn't currently collecting, so the only thing lit up is where the hunt actually is.
• Route recording — hit record and your drive paints onto the map at road width with direction arrows, so you can tell which side of the street you've covered. Recording survives the screen going off or the phone sitting in a pocket, and if the app gets killed mid-session it resumes on relaunch.
• Persistent route history — every past route stays painted on the map across sessions, with a choice of display modes, so you can see at a glance what you haven't hit yet.
• Pins for good finds — long-press to drop a numbered pin when you spot something but don't have room in the car, so you can swing back later.
• Session stats and playback — distance, time, average speed and an estimated fuel cost per session, plus all-time totals, and a route-playback screen to scrub back through a past drive.
• Practical extras — a public toilet layer for long hunts, suburb star ratings, an animated first-run onboarding tour, and a Smart Rate GPS mode that eases off sampling at speed and sharpens it at low speed to save battery.
• Privacy-first by design — no accounts, no sign-up, no ads. Everything lives on-device first. An anonymised, trimmed copy of each route (first and last ~300m cut, anything under 600m excluded entirely) uploads to power a future community heatmap, with no user or device identifiers attached.

Under the hood

KerbQuest exists as three clients sharing one backend, though the iOS app is the shipping, actively-developed product. The native app is SwiftUI and MapKit, around 4,100 lines of Swift across a clean feature-module layout, Map, History, Onboarding, Settings, Persistence, Services, Networking, Design System. I chose MapKit over Mapbox deliberately, for tighter native integration and better offline behaviour, and this is the version going to the App Store.

The original build was a React/TypeScript/Vite web app, which now serves as a functional web fallback of the app itself at kerbquest.com, plus an info site at kerbquest.com/app. There's also an earlier-stage Expo/React Native build. Underneath all three sits an Express/TypeScript backend on Railway for session sync and the Council's clearout schedule, plus a Cloudflare Worker piping anonymised, trimmed route uploads into an R2 bucket, laying the groundwork for a future opt-in community heatmap of collection hotspots.

I conceived, architected, directed and debugged the whole build, with substantial AI assistance from Claude on implementation. Working that way, knowing exactly what I want built and why, then using AI tools well enough to actually ship a real native app with it, is as much the point of this project as the app itself.

Decisions worth mentioning

• Design system over defaults — a hand-built brand palette, deep greens and warm accents, applied consistently across onboarding, map UI and controls, a deliberate move away from generic styling.
• Onboarding as a feature, not an afterthought — splash, location-permission primer, then an optional animated tour with feature badges anchored to real UI positions, built to convert first-time installs rather than dumping people straight onto a map.
• Privacy engineering over privacy theatre — the telemetry trimming was a specific technical decision to make it structurally impossible to infer a user's home address from uploaded routes, done before App Store privacy-label submission, not bolted on after.
• Resilience for the core use case — since the app's whole job is "record a drive," session recovery across app kills, backgrounding and screen-off got real engineering attention rather than being treated as an edge case.

Where it's at

KerbQuest is in TestFlight beta now, ahead of a public App Store submission. Submission copy, keyword research and the privacy nutrition-label mapping are all prepared and checked against the actual shipping code, not aspirational feature copy. A paid tier, community heatmap, CarPlay, iCloud sync, gamification, coverage beyond Brisbane, is scoped but deliberately deferred until there's real usage signal from the free release.

The stack

• Swift
• SwiftUI
• MapKit
• Combine
• TypeScript
• React
• Vite
• Tailwind CSS
• shadcn/ui
• Mapbox / Leaflet
• Expo / React Native
• Node.js
• Express
• Drizzle ORM
• Cloudflare Workers + R2
• Railway

Screenshots