Back to case studies

VoiceDesk AI Voice Receptionist

Voice AIAI AgentsMulti-Tenant SaaS

A multi-tenant SaaS platform that lets any service business — healthcare, legal, real estate, automotive, home services, and more — deploy an AI voice receptionist trained on its own services, FAQs, pricing, and booking rules, then reach customers through a website widget or a phone line with real-time, human-like conversations.

Try the live demo

Key Capabilities

  • Real-time voice over OpenAI GPT-4o Realtime sessions, reachable from a one-script-tag website widget or a Twilio phone number, with sub-second responses, natural interruptions, and multi-turn context.
  • Agent orchestration: a receptionist agent determines intent and hands off to booking, lead-qualification, FAQ, and knowledge agents through OpenAI function calling, with every function call validated before execution.
  • Appointment booking that checks availability, books, reschedules, cancels, and confirms against Google Calendar and Outlook (Microsoft Graph), keeping every appointment in sync with the calendar provider.
  • Lead qualification that captures name, contact details, requested service, budget, and timeline, then syncs the lead to HubSpot or Salesforce.
  • Per-tenant knowledge base built from FAQs, service catalogs, pricing, policies, hours, and uploaded documents, stored as vector embeddings so answers stay grounded in each business's data.
  • Conversation intelligence: every call produces a transcript, summary, sentiment, intent, and conversion score via background Inngest jobs, feeding an analytics dashboard of calls, bookings, leads, and conversion.
  • Multi-tenant security: every record and query is organization-scoped, five-role RBAC, 15-minute JWT access tokens with rotating refresh tokens, Argon2 password hashing, and encrypted, access-controlled call recordings.

Tech Stack

Next.js 15TypeScriptOpenAI GPT-4o RealtimeTwilio VoicePostgreSQL (Neon)PrismaInngestUpstash RedisGoogle Calendar APIMicrosoft GraphStripePostHogVercel

Interested in something like this?

Let's talk