System architecture, backend, configuration UI, integration design
System
Node.js, Express, Google Calendar, Redis, VAPI, Retell, Dentin
Service qualification rules for a home-services voice agent.
01
Voice infrastructure
A useful voice agent has to do more than hold a conversation. It must understand whether a job is supported, confirm service coverage, select the right calendar, offer valid times, and carry a chosen slot through to booking.
TARK AI provides that operational layer. The language model handles the conversation while the backend turns business rules and live systems into dependable actions.
02
Two workflows
Home services
Each tenant configures unit types, subtypes, brands, residential or commercial work, service areas, technician calendars, working hours, and scheduling zones.
Dental clinics
Clinic configuration maps procedures to appointment types and providers. Runtime calls expose short slot identifiers so the voice agent never has to reason about internal provider or operatory IDs.
03
Runtime path
The workflow narrows uncertainty before any appointment action is attempted.
01Call context
02Qualification
03Coverage
04Availability
05Slot token
06Booking
04
Production control
The architecture gives voice agents freedom in conversation while keeping serviceability, provider routing, slot selection, stale-slot recovery, and booking deterministic.
That separation turns a flexible AI conversation into a system that can make real operational commitments without losing control of the transaction.
Qualify the request before spending time on scheduling.
Keep credentials and integration logic behind the API.
Recover explicitly when availability changes or booking fails.
Capture inbound and outbound activity for production diagnosis.