Face recognition attendance

Context

An event staffing company wanted to replace badges and paper lists with face recognition, paying each person for the time they actually worked. The pilot ran in a stadium, on unstable networks and ordinary devices.

What I did

• On-device face detection and face embeddings with Google ML Kit and TensorFlow Lite, processing camera frames in real time with CameraX.
• Kotlin with Clean Architecture (MVVM, Hilt) and a Jetpack Compose UI.
• Integration with a Python backend on AWS through multipart uploads, retries keyed by tracking ID for unstable networks, and Sentry monitoring.

Outcome

Shipped and used in the pilot. The client paused the project after that phase.