RIPPLE

Infrastructure issues go unreported because the friction to report is too high. Council portals require accounts, forms, and manual categorization. Without volume data, councils can't prioritize repairs effectively. Communities have no way to signal which issues matter most.

Built a 3-second reporting pipeline: snap a photo, on-device AI classifies it, geolocation pins it to a community map. Community upvoting creates social proof for council prioritization. Zero accounts, zero personal data, all AI runs on-device.

Ripple started from a simple frustration: you see a pothole every day on your commute, report it to the council, and nothing happens. Multiply that across a city and you get thousands of unreported issues because the friction to report is too high and councils have no way to prioritize.

Ripple makes reporting take 3 seconds. Open the app, snap a photo, and the on-device AI (TensorFlow.js running MobileNetV2) classifies the issue · pothole, broken streetlight, graffiti, accessibility hazard, illegal dumping, and 5 more categories. The report is automatically geolocated and pinned to a live community map. No typing, no forms, no account required.

The community layer is what makes it work at scale. Neighbors see reports on the map and upvote issues they encounter too. When a pothole has 47 votes, the council knows it's not one person complaining · it's a neighborhood consensus. The live Mapbox map uses clustering, category-colored pins, and heatmaps to visualize problem density across suburbs.

Pilot deployment targets 5 Melbourne councils: City of Melbourne, City of Yarra, Moreland, Darebin, and Port Phillip. Privacy-first: no accounts, no personal data collection, all image classification runs on-device.

Frontend is React 19 with TypeScript, Vite, and Tailwind CSS, built as an offline-first PWA. Mapbox GL JS 3.20 renders the dark-v11 base map with category-colored pins, clustering at zoom levels, and heatmap overlays for issue density.

The AI pipeline runs entirely client-side. TensorFlow.js loads MobileNetV2 on app init and classifies photos locally · no images are ever transmitted to a server. Classification results (category + confidence score) are sent with the geolocated report to Supabase.

Backend runs on Supabase: PostgreSQL with Row Level Security, Realtime channels for live report updates, and Deno Edge Functions for aggregation. Elasticsearch provides full-text search across reports and powers analytics dashboards for council operators. Resend handles transactional email notifications.

Offline support uses IndexedDB (via IDB library) to queue reports when connectivity drops. Reports sync automatically when the device reconnects. Zustand manages client state with persistence.