ProFynd.
Geolocated tutor-matching application. Android, Google Maps API, Firebase. Interest-based pairing and in-app messaging.
Match quality decays gracefully.
Android-first. Google Maps for geolocation, Firebase for backend and messaging. The pairing function is subject interest × confidence-weighted distance — close-by matches dominate when there are many candidates, distance loosens as the candidate set thins. The decay is intentional: outside the dense city centre, you'd rather a 4 km match in your subject than a 200 m match outside it.
In-app messaging is Firebase Cloud Messaging end-to-end; no third-party chat SDK. The decision saved a service-level dependency and let us keep the cold-start latency under 800 ms on mid-range devices.
What's still standing.
ProFynd is no longer actively maintained — it was a final-year undergrad project that taught me Android, Firebase, and the importance of a confidence score on every retrieval. The repository is public.
Gallery.
The harness teaches you what the model is doing. Build it first; the rest is implementation.