The book walks through building FinRisk, a credit risk app that runs a trained model on-device in <15ms. The model is 1.8 KB. You train it in Python, convert to TFLite, and integrate into an Android app with Clean Architecture + Hilt + Compose.
Chapter 1 is free: https://7422889060748.gumroad.com/l/ml-android-preview
Source code is open: https://github.com/vsay01/FinRisk
Happy to answer questions about on-device inference, TFLite, or the Android integration.