Supercharging Android Apps with On-Device AI: Gemini Nano & Prompt API
Mobile AI is moving on-device, driven by the need for privacy, low latency and offline use. This session shows how to build intelligent Android apps that run entirely on the phone with Gemini Nano, the ML Kit Prompt API, the ML Kit GenAI APIs and the MediaPipe LLM Inference API.
The centrepiece is a real-world demo: an AI-driven activity booking search built on Gemini Nano with the Prompt API, where prompt design and best practices cut response time from 18 seconds to 3. From there it covers what it takes to design, optimise and ship on-device AI features in consumer and enterprise apps.
Key takeaways
The on-device AI landscape
Why AI is moving from the cloud to the device, and how Android abstracts it through AICore, LoRA fine-tuning, Private Compute and NPUs.
Getting started with Gemini Nano
Integrating Gemini Nano in a modern Android app through the ML Kit GenAI APIs and the Prompt API.
Prompt API in production
Designing and optimising prompts for an AI booking search, from 18 s to 3 s per answer.
Production considerations
Model size, device compatibility, privacy and performance lessons from shipping AI features at scale.
Beyond Gemini
MediaPipe, LiteRT and custom models for specialised on-device tasks.