Portfolio
Inbound Calling Platform with LiveKit, Custom STT, and LLM Orchestration
Routes live calls into a low-latency pipeline that transcribes audio, detects intent, orchestrates LLM responses, and generates call dispositions.
Project details
What we delivered
Overview
Production inbound calling platform that processes live calls and generates structured call outcomes.
Supports real-time transcription, intent detection, smart routing, and post-call reporting.
Key Features
- Live audio bridging with LiveKit for low-latency sessions.
- Custom STT pipeline optimized for call center acoustics.
- Real-time intent detection with LLM orchestration for context-aware answers.
- TTS fallback and agent-assist suggestions for live responses.
- Automated call disposition reporting with duration, intent, and outcome.
Architecture & Tech Stack
- Media transport: LiveKit sessions for call audio.
- Speech stack: custom streaming STT with pre and post processing.
- LLMs: Vertex AI or in-house models with prompt engineering for intent and summaries.
- API and orchestration: FastAPI services with session control and monitoring.
- Analytics: time-series database and BI dashboards.
Impact / Outcomes
- Converted live calls into structured data for downstream automation.
- Improved disposition accuracy and reduced time to campaign reporting.