Professional work
AI Voice Orchestration
Real-time voice agents for retail and hospitality, orchestrated as LangGraph state machines that call tools and retrieve context mid-conversation.
- role
- Lead Engineer
- org
- Rezolve AI
- when
- 2025 – Present
PythonFastAPIPipecatLangGraphLiveKitWebRTCAzure SpeechRedisAKSHelmMCP
01
Overview
An enterprise Voice AI platform for conversational experiences across retail and hospitality. I lead its design and development, with an emphasis on low-latency performance and a cost-efficient architecture.
02
Architecture
- 01
Caller
Browser or app, streaming audio over WebRTC
- 02
LiveKit
Real-time media transport
- 03
Pipecat pipeline
Azure STT → agent → Azure neural TTS
- 04
LangGraph agents
Multi-agent flow with tool calling and RAG
- 05
Business systems
Reached through MCP connectors and SDKs
03
What I built
- Real-time voice pipeline with Python, FastAPI, Pipecat, LiveKit, WebRTC and Azure Speech Services
- Modular multi-agent architecture combining LLMs, tool calling, retrieval-augmented generation and external services
- Reusable SDKs and MCP-based connectivity so agents can securely reach external tools and customer platforms
- Cloud-native deployment on Azure Kubernetes Service with Docker, Helm and CI/CD pipelines
04
Challenges
- Keeping end-to-end latency low enough for conversation to feel natural
- Speech-recognition accuracy on real-world audio
- Reliable audio streaming and graceful handling of user interruptions