Building Privacy-First Android Apps with On-Device AI Models

Most AI applications rely on cloud services, raising concerns around privacy, latency, and cost. This session demonstrates how Android developers can build privacy-first applications by running AI models directly on-device.

The session focuses specifically on AI models officially supported for on-device usage in Android, using tools like TensorFlow Lite, MediaPipe, and on-device capabilities like Gemini Nano (available on supported devices).

It will also cover how to evaluate whether a model is suitable for on-device execution based on factors such as model size, memory usage, hardware acceleration (CPU/GPU), and inference latency.

A live demo will showcase integrating and running an on-device AI model in an Android application, highlighting real-world performance and usability.

1. Understand on-device AI and how it differs from cloud-based AI

2. Learn how to identify and use officially supported on-device AI models in Android

3. Understand how to evaluate model suitability based on device constraints (size, memory, performance)

4. Gain practical experience integrating on-device AI using TensorFlow Lite and MediaPipe

5. See a real-world demo of a privacy-first Android app powered by on-device AI