Put your phone in airplane mode. Now ask your app a question—and watch it answer.
In this hands-on session, we’ll build an AI-powered Kotlin Multiplatform app that runs Gemma models directly on Android and iOS. No API keys, no inference server, and no internet connection once the model is downloaded.
We’ll follow the path from choosing a model that fits in your pocket to streaming its first response. Along the way, we’ll explore what belongs in shared Kotlin code, what needs a platform-specific implementation, and how to connect native inference runtimes behind a common API.
Then we’ll tackle the details that make or break the experience: large downloads, slow model loading, memory limits, battery use, and devices that perform very differently from your development phone.
Expect live demos, real Kotlin code, and an honest look at the trade-offs. You’ll leave with a repeatable approach to bringing offline AI into your own KMP apps.
Learning outcomes
- Choose a Gemma model and quantization appropriate for mobile devices.
- Design a shared Kotlin API around platform-specific inference engines.
- Stream model responses into a Compose Multiplatform UI.
- Manage model downloads, loading states, and resource lifecycles.
- Measure performance and understand the trade-offs on real devices.