# VeraVoice — AI Phone Receptionist

Canonical URL: https://alectorlab.com/work/veravoice-ai-receptionist
Raw Markdown: https://alectorlab.com/work/veravoice-ai-receptionist/raw
Industry: Customer Experience / Small Business
Capabilities: Agentic AI, AI Product Engineering
Owner: Alector Lab – Enabling Technologies Ltd (direct)
Provenance: Alector Lab project record; operational metrics are not independently audited.

## Summary
A 24/7 AI receptionist that answers every call instantly, books appointments, and screens spam — so a busy business never loses a customer to a missed call again.

## Operational Problem
Missed calls cost service businesses bookings: without round-the-clock coverage, calls outside business hours go to voicemail or the next business that answers.

## Operating Context
VeraVoice is built as a phone-answering agent rather than a scripted IVR menu — it holds a conversation, understands what the caller wants, and takes action on it directly.

## Engineering Approach
The agent runs on OpenAI's Realtime API for conversation, routed through Twilio for telephony and LiveKit for low-latency audio streaming, backed by a Node.js service that handles appointment booking, spam screening, and call logging.

## System Architecture
1. Telephony: Twilio handles inbound call routing and number provisioning
2. Voice Agent: OpenAI Realtime API for conversation
3. Media Streaming: LiveKit for real-time audio transport
4. Backend: Node.js service handling appointment booking, spam screening, and call logs

## Implementation Details
The agent answers every inbound call, determines caller intent, books appointments directly into the calendar, and filters spam and robocalls before they reach staff.

## Governance-Reviewed Outcomes
- Qualitative workflow record; numerical performance metrics are withheld unless formally signed off for public release.

## Technology Stack
- Twilio
- OpenAI Realtime
- LiveKit
- Node.js

## Engineering Lessons Learned
- A phone agent has to handle interruptions and natural speech, not just clean turn-taking, to be usable as a real receptionist.
- Spam and robocall filtering needs to run before the conversation flow starts, so it doesn't waste the agent's time or clutter the call log.