16:20 ~ 17:00 Meerkat
Server-Sent Events on Android: Real-Time Streams That Survive the Factory Floor 2026.09.02 16:20 ~ 17:00 Otter
Android Gets a Voice: Building Real-Time Agents
Android Platform English Interpretation available
Mark WickhamM15.ai, CEO
Most Android voice implementations are still turn-based: tap, speak, wait, respond. This talk shows how to break that paradigm and build a real-time, continuously interactive voice agent on Android that listens, thinks, and speaks simultaneously—closer to human conversation than command execution. We'll walk through a production-grade architecture built with Jetpack Compose and a foreground service for continuous audio capture, integrating streaming speech-to-text, an LLM, and low-latency text-to-speech. Both cloud and on-device approaches are covered, with honest tradeoffs around latency, cost, and privacy. Latency is the single biggest design constraint in voice systems, so we'll spend focused time on where it hides: audio buffer sizing, endpoint detection tuning, network jitter, token streaming strategies, and playback queue management. Attendees will leave with a latency profiling framework and concrete per-stage budgets they can apply to their own architectures. We'll also tackle the hard problems usually glossed over: barge-in, echo cancellation, and graceful interruption. A voice agent needs a voice-native UI, so we'll build an audio-reactive visualizer in Jetpack Compose—an animated orb that responds to mic amplitude, LLM processing state, and TTS playback in real time, giving users a continuous sense of full-duplex conversation. Users can toggle between the visualizer and a traditional scrolling transcript. We'll also cover securing the pipeline for production: managing secrets with Android Keystore, encrypted SharedPreferences, and server-side proxying—because hardcoded keys ship more often than anyone admits. As a unifying demo, we'll build an complete voice interface that brings all the pieces together. Attendees get working, production-ready reference code they can adapt directly. The session will also explore how these building blocks can be integrated in Open Claw agentic pipelines. By the end, you'll know how to design and ship real-time voice agents on Android—from audio pipeline to animated UI to secure deployment.
Intended audience
Android engineers familiar with Kotlin and Jetpack Compose who want to build real-time voice interfaces. Comfort with foreground services and audio APIs is helpful but not required.