¿Bueno, me escuchas? Building Voice Agents With Livekit

​A hands-on workshop. Bring your laptop, leave with a working voice agent.

​In this workshop we build a real-time voice agent on LiveKit, step by step, together. It's not a talk: everyone writes code on their own machine, and by the end of the session you have an agent that picks up, listens, answers, and handles being interrupted mid-sentence.

​We start with the anatomy of the system — speech-to-text, an LLM in the middle, text-to-speech, and the orchestration that makes it feel like a conversation instead of a walkie-talkie. From there we start from a working template and build it up piece by piece, with pauses so nobody gets left behind. The last stretch is open: change its personality, wire up a tool, swap the voice, whatever you want.

​Who this is for: engineers who want to ship voice, not just read about it. You should be comfortable in Python or TypeScript. No prior voice or WebRTC experience needed — the interesting problems here are latency, turn-taking, and interruption handling, and none of them require you to know what a jitter buffer is.

​What you'll walk away with

​A voice agent running, yours, in your own repo An understanding of where latency goes in a voice pipeline and how to cut it How to handle interruptions and turn-taking without it sounding robotic A decent starting point for taking it to production