Project
CrocSpy
From a mobile app to a research PWA.
A UFBA research project exploring how photos and location could help identify and catalog crocodilian species. The app became a public PWA; the planned AI service did not reach deployment.
Public research prototype
Demo classifications are simulated. The AI model was not deployed as a service.
My role
I led the app development team. Other students handled AI training and the service intended to connect the model to the app. My contribution centered on the app experience and the move from React Native to a web app.
From React Native to PWA
Publishing costs were the main reason we changed direction. A PWA could be distributed through a web link, with local storage and cached pages supporting offline access to the app. That offline support covers the interface and local data, not AI inference.
Inside the PWA
- React and Vite PWA, following an initial React Native app.
- IndexedDB through Dexie for local storage; a service worker caches the app pages and assets.
- Leaflet for the map and browser camera access for photo capture.
What shipped, and what remained blocked
The public prototype brings photos and locations into a collection that can also be explored on a map. Both the university deployment and the alternative Vercel deployment are linked above.
The AI team did not receive the university computing allocation needed to deploy the model service, so we could not complete that integration. The demo simulates uploads and assigns species randomly: these are not model predictions and should not be used to identify an animal.