Real-time conversational voice AI engineer (ElevenLabs, WebRTC) — fix orb states + latency

I'm building an AI language-learning app where users have spoken conversations with an AI tutor shown as a voice "orb" (beginner/intermediate/advanced levels and roleplay). I need an experienced real-time conversational voice AI engineer to diagnose and fix specific issues. This is in-app voice over WebRTC/WebSockets — NOT phone or telephony call agents, and NOT general app development. Two problems: (1) the orb's "listening" and "talking" states glitch and don't switch cleanly, and (2) noticeable latency, plus occasional AI hallucinations that drift off the lesson. Must have hands-on experience with: ElevenLabs Conversational AI, LiveKit or Pipecat, real-time turn-taking and barge-in handling, and RAG for grounding. Please share a link to a LIVE in-app voice product you've built (not just a demo). Starting with a small paid diagnostic — a few hours to reproduce the issues and tell me the root cause and your fix — with ongoing work likely after.