Machine Learning for Android Engineers: From Theory to On-Device Inference

You're an Android developer. Every job posting now mentions "ML experience preferred." Your PM keeps asking about AI features. But every tutorial either drowns you in math or gives you copy-paste code you don't understand.                                                                                                    

This book builds your mental models and engineer intuition — not memorization, not math. Every concept is explained through analogies you already understand as a developer, so you know why things work, not just how to copy them.                                                                                                  

You'll build FinRisk, a complete credit risk app that runs a trained ML model entirely on-device. No server, no API calls, no internet required. The final models are 1.8 KB and 2.6 KB. Smaller than most app icons.                                                                                                                        

What's inside (134 pages, 6 chapters):                                                                         

- How models learn: training loops, loss functions, gradients, and backpropagation — built on intuition, not equations                                                                                                      

- End-to-end build: Python training, TFLite conversion, Android integration with Clean Architecture + Hilt + Compose                                                                                                        

- Upgrade from logistic regression to a neural network — without changing a single line of Android code        

- Production hardening: crash reporting, fallback classifiers, kill switches, model monitoring, drift detection                                                                                                              

For Android developers (2+ years) who want to add ML features without becoming data scientists. Not for LLMs/ChatGPT — this is on-device inference, models that run locally with zero network dependency.              

Complete source code available at github.com/vsay01/FinRisk