Real-time AI receptionist with full observability
LiveKit Doctor Receptionist
LiveKitGeminiElevenLabsDeepgramLangfuse
Problem
Medical front desks need a receptionist that can hold a natural phone conversation, manage appointments, and stay reliable — with every interaction traceable for debugging and quality review.
Solution
Built a real-time AI receptionist using LiveKit agents, Google Gemini for reasoning, ElevenLabs for TTS, and Deepgram for STT. Integrated Langfuse for tracing, logging, and prompt evaluation. The agent manages appointments, refreshes patient data, responds conversationally, and handles fallback logic with full traceability.
Tech Stack
- LiveKit — Real-time audio transport and agent orchestration for the call.
- Google Gemini — Conversational reasoning and appointment-handling logic.
- ElevenLabs — Natural-sounding text-to-speech for the receptionist's voice.
- Deepgram — Low-latency speech-to-text for caller input.
- Langfuse — Tracing, logging, and prompt evaluation for production observability.
Screenshots and a demo walkthrough are coming soon.
Challenges
Write-up coming soon.
Impact
Metrics coming soon.
Lessons Learned
Write-up coming soon.