Talk to it. It talks back. Abdalla builds the pipeline where you speak, a model thinks, and a voice answers before the awkward silence becomes a feature. No hold music. No “your call is important to us.”
Self-hosted STT and TTS. LiveKit in the middle. The kind of stack that hates lag and has feelings about packet loss.
Speech in, speech out
STT → LLM/agent → TTS. Streaming audio. WebRTC. Abdalla has spent more time than is healthy making sure the first word does not arrive next week.
Rooms that stay open
LiveKit, MediaDevices, ICE, STUN/TURN. If it needs a real-time conversation — brokers, avatars, the 2am WhatsApp voice — it lives here.