Architecture Lead

AI Voicebot Architecture

Designed AI-powered customer-support voicebot architecture involving real-time audio streaming, STT/TTS, multi-agent orchestration, SOP execution, backend tool calls, and observability.

AILLMVoicebotLiveKitWebRTCAutomation

Context

Customer-support calls often involve repeatable order, return, and pickup-related queries that can be automated with safe AI workflows.

Problem

Traditional IVR systems are limited for multi-turn conversations and do not handle dynamic support workflows well.

What I Designed

WebSocket-to-WebRTC audio flow.
LiveKit-based call sessions.
STT/TTS voice pipeline.
Multi-agent orchestration.
SOP/state-machine execution.
MCP-based backend tool calls.
Caching and observability.

Technical Highlights

Real-time audio streaming
Session handling with LiveKit
State-machine based SOP execution
Backend tool-call integration
Latency and observability planning

Impact

Created architecture for AI-powered customer-support automation.
Enabled structured handling of order, pickup, return, and escalation flows.
Improved clarity around latency, session handling, and backend integration.

Skills / Tags

AI Automation
LLM Integration
LiveKit
WebRTC
STT/TTS
Observability