WarsawJS: JavaScript Community Meetups in Warsaw

Daniil Rabizo

State of Voice AI with JS 

Voice agents feel magical until someone talks over them. This talk starts with a working agent built on LiveKit, then takes it apart: how the pieces fit together, and where the whole thing tends to fall over. We'll go through latency — where the milliseconds actually go across STT, LLM, and TTS, and which ones you can buy back; interruptions — barge-in, turn detection, and what to do with a half-spoken sentence; stability — how to move fast with agent development; and guardrails — keeping an agent on topic when the input is unscripted human speech. A live demo, an architecture walkthrough, and the mistakes you don't need to repeat.