Beyond the Chatbox: Ship On-Device AI in Your Browser

Most AI features look the same: a chat panel bolted onto the side of the app. Chat is the command line of the AI era, powerful but lazy. In this lab, you'll build the alternative with your own hands: real ML models running inside the browser, powering interfaces that see, suggest, and react in real time, with no API key, no round trip, and no user data leaving the device.
We start on familiar ground: a Spring Boot + Kotlin service running pose estimation on the JVM with DJL. You'll measure its round-trip latency and write the number down. Then we move the model into the browser with LiteRT.js and WebGPU, step by step: first inference on a static image, then a live camera feed with a real-time skeleton overlay. In the final act, you'll race your browser against your own backend and produce the latency comparison that explains when AI belongs on-device, in the cloud, or nowhere at all.
You'll leave with a working project on your laptop, two bonus labs (in-browser semantic search and a camera-driven form that fills itself) to continue on the train home, and a decision framework you can defend in your next architecture discussion. The GDPR argument for inference that never leaves the device comes free with the code.
No machine learning background required. If you can build a Spring service and read TypeScript, you're good-to-go.