FaceMatch / WIMS

FaceMatch / WIMS is a mobile enrollment and recognition product: on-device face detection, liveness checks, and recognition pipelines, with optional integration to a WIMS-style backend for device registration, heartbeats, queued work, and syncing recognition results when connectivity allows.

My role

I worked across the Android (Kotlin + native C++ via JNI) and iOS (SwiftUI) clients, plus a small Python FastAPI harness used to exercise the mobile API during development. The emphasis was on camera-first UX, reliable persistence of enrollment and attendance rows, and operational visibility (settings, logging) for field use.

What we built

• Performance-sensitive vision on Android by keeping heavy work in native code while the Kotlin layer handles UI and lifecycle.
• Cross-platform parity for the same product story: enrollment flows, recognition, and background behaviour aligned between Android and iOS where the platforms allow it.
• Resilient sync: local rows with status fields so failed uploads can retry instead of silently dropping data.
• Credential handling using platform-appropriate secure storage instead of plain preferences for tokens and device keys.

Stack

Kotlin, C++ and CMake, Android CameraX, Retrofit, Room, WorkManager; Swift, SwiftUI, TensorFlow Lite on iOS; OpenCV- and TFLite-oriented models in assets; Python FastAPI for local API mocking during development.

What made it hard

Camera pipelines and concurrency do not forgive sloppy threading—especially on iOS where actor isolation and video capture queues intersect. The work was as much about correct frame handling and settings sync as it was about model thresholds: the operator has to trust what the device shows them in real time.