01 // ML · ANDROID
Digital Behaviors
A privacy-focused mobile application analyzing device-use patterns and transforming screen metrics into actionable behavioral insights.
TELEMETRY AT A GLANCE
- Native Android application architecture
- On-device Machine Learning inference (TFLite)
- Zero server-side telemetry or privacy leakage
System Overview
This project was engineered to make digital habits transparent and actionable. Instead of treating screen time as an aggregate scalar number, the system captures a multidimensional view of device interaction frequencies, volatility cycles, and focus endurance.
The Core Problem
Standard OS screen-time monitors fail to account for context or cognitive fragmentation. Users lose track of how frequently they context-switch between applications, resulting in productivity drop-offs without clear diagnostic insights.
Technical Architecture
Android Telemetry Collector
Feature Extraction Pipeline
TensorFlow Lite Inference Engine
Behavioral Diagnostic UI
Technologies & Toolkit
- Android SDK & Kotlin Coroutines
- TensorFlow Lite on-device model quantisation
- Scikit-learn & Random Forest Classifiers
- Local Room database with end-to-end encryption
Engineering Challenges Overcome
- Minimizing background battery drain while periodically computing behavioral telemetry vectors.
- Quantizing ML models to run with sub-10ms latency on resource-constrained mobile hardware.
- Designing an intuitive, high-contrast dashboard that highlights focus trends at a single glance.