Building an AI companion with Modern Android Architecture
summary
Builds Microsoft Copilot for Android from scratch using Jetpack Compose, Hilt, Retrofit and Kotlin serialization; ships a modular answer cards framework that isolates UI/data per feature and uses Kotlin polymorphic serializers plus multi‑binding into map/set to load converters at runtime. Uses a convention Gradle plugin to standardize modules, feature flags and client capability flags over a WebSocket for progressive rollout, and a default serializer fallback for unknown cards. Timeline: first PR April 2024, public October, scaling to 12M active devices while maintaining a 99.99% crash‑free rate.
AI-written summary · about
chapters (15)
- 0:00Building an AI companion
- 1:05Copilot features overview
- 2:30Project timeline and scaling
- 3:35Merge conflicts in chat
- 6:45Gradle setup with plugins
- 8:12Answer Card framework
- 11:41Finance cards implementation
- 16:43Data layer design
- 20:25AI responses to UI models
- 23:17Finance card demo
- 25:02User interactions in cards
- 26:28Safety with feature flags
- 29:19Modular development strategies
- 30:01Results and closing
- 31:28Q&A
related talks
33mCopilot in Android: AI for boring stuff, people - for important
41mKeynote: Now in Modern Android
29mAndroid Gets a Voice: Building Real-Time Agents
58mModern Android Development
35mNext Generation Android UI with Jetpack Compose
51mModern Android Development
39mWhat's new in Android Studio's AI Agent?
34mRetrofitting Legacy Code with Jetpack Compose at Scale