This session is a retrospective on the engineering behind a pocketwatch that runs a real time voice agent. Built end to end in a single week with Claude Code, from product idea to working reality complete with its own product page, this project showcases an entirely agentic workflow for rapid prototyping and iterative development.

We will explore the ESP32 S3 audio pipeline, including capture, echo cancellation, a unified STT LLM TTS agent, and the barge in system that lets users interrupt mid sentence. You will see how the device’s own settings such as display mode, brightness, volume, color themes, and even its name can be changed through natural speech.

With a fully animated AMOLED UI and synchronized playback, this project demonstrates how conversational interfaces can feel personal and immediate on microcontroller class hardware. Attendees will leave with practical patterns for building voice agents that can manipulate their own environment.