WSF
An AI safety system that acts before anyone presses SOS.
- Year
- 2026
- Role
- Location tracking and data layer
- Team
- Team
- Stack
- FlutterFastAPISupabasePostGISscikit-learnTFLiteNext.jsMapbox
01 — The problem
A women's safety platform that learns where incidents happen, routes people around danger and escalates on its own when something goes wrong, even if the phone is out of reach.
Most safety apps are a button. In the moments that matter, pressing it is often impossible.
WSF watches, routes and listens, and acts on its own when the signals say something is wrong.
02 — What it does
Threat zones
Incident points are clustered with DBSCAN and wrapped in concave hulls stored in PostGIS. Zones shift with time of day and severity.
SafeNav routing
Routes that pass within 50 m of a red zone are flagged, and the safest of the fastest three is chosen.
Drift protocol
Leave the safe route by more than 50 m for 30 seconds and the app checks in silently. No answer means an automatic SOS.
Edge audio sentinel
YAMNet runs on the phone and listens for screams, impacts or breaking glass. Audio never leaves the device.
03 — How it works
- 1
Flutter mobile app, FastAPI backend, Supabase with PostGIS for auth, real-time location and geospatial zones, and a Next.js + Mapbox command dashboard for dispatchers.
- 2
Battery-first tracking hands location to the native OS geofencing layer, so the CPU only wakes every ten metres.
04 — My part
- Rebuilt the location-tracking pipeline on top of native background geolocation.
- Unified the Supabase schema so the app, the backend and the dashboard share one source of truth.
ISL Recognition
Real-time Indian Sign Language recognition in the browser.