Gen AI for Android Developers
summary
Introduces an on-device GenAI stack for Android centered on Gemini Nano V3 running on Pixel 10, plus an AI Core runtime that manages model download, storage and per-chip calibration, and ML Kit GenAI with an experimental Prompt API for apps. Shows a KakaoT demo that extracts sender/recipient names and phone numbers with about 96% accuracy using prompt engineering and few-shot examples instead of ~1,000 labeled training cases. Notes support for Gemma 3n, ICL/LoRA fine-tuning, LiteRT Next and Play Delivery, and that the stack is currently enabled on 33 devices.
AI-written summary · about
chapters (9)
- 0:00Gen AI for Android intro
- 1:26Why Gen AI is needed
- 4:19On-device Gen AI
- 7:12Gemini Nano evolution
- 11:45How Gemini works
- 22:02Prompt engineering
- 33:00Prompt optimization tips
- 37:19Use cases and success
- 38:46Magic Suggest outlook
related talks
41mBuild intelligent Android apps with Google's AI
30mGenAI on Android with Gemini Nano and Firebase
39mAdd Generative AI capabilities to your Android app
10mAI in your pocket: Building intelligent Android apps
39mGetting Started with GenAI for Android Apps
21mAdd Generative AI capabilities to your Android app
23m2025: On‑Device AI Goes Mainstream on Android
48mAndroid + AI: Solving Problems Smartly