AI for Android on- and off-device
summary
Live demos show an Android app using Onnx Runtime for on-device inference (a 12 MB object detector that misses some statues) and a local build of Whisper (about 57 MB) plus cloud fallbacks to OpenAI GPT-3.5 for summarization and chat. A simple text editor sends HTML to OpenAI and returns nondeterministic summaries; a JetChat client uses GPT with a mutable conversation buffer. Conference schedule is embedded as 1,500‑dim vectors and matched by dot‑product with a 0.8 cutoff to ground answers, and hybrid local/cloud fallback logic is implemented and functional.
AI-written summary · about
chapters (5)
- 0:00Introduction and demos
- 8:43AI models cloud vs device
- 19:10LLM APIs completions chat embeddings
- 37:15On-device models and runtimes
- 42:28AI safety and user experience
related talks
10mAI in your pocket: Building intelligent Android apps
20mAI in Android Development at Duolingo
24mMake your app smarter
41mGen AI for Android Developers
23m2025: On‑Device AI Goes Mainstream on Android
32mAI for Mobile and IoT Devices: TensorFlow Lite
36mMaking intelligent Android apps
20mAI and Machine Learning for developers