Project Overview
This is a real-time conversational voice agent designed for customer-support automation. It runs on LiveKit WebSockets, streams speech-to-text through AssemblyAI, generates replies with a large language model, and speaks them back with Cartesia Sonic text-to-speech.
The subtle-but-critical piece is Silero VAD (voice activity detection), which handles accurate turn detection — knowing when the caller has finished speaking. Get that wrong and the agent talks over people; get it right and the conversation feels genuinely human.