Home/Work/AI receptionist platform for clinics and practices
Healthcare · Conversational AI · SaaS · Case study

An AI receptionist a clinic can configurein an afternoon and trust withits calendar.

A multi-tenant receptionist on voice, WhatsApp and web: onboarding compiles a working agent from a business's answers, bookings pass a server-side guard, and a test lab blocks bad prompts.

3channels, one agent per tenant
1session to configure an agent
0double-bookings by design
100%of prompt changes tested before deploy
Inbound call · Northgate Dental live
Slots offered2
Guard checks3
Handovers0
Happening now
  • find_slots(fri, pm)2 results21ms
  • booking_guard(2:30)no overlap
  • book(2:30) · release(thu)committed34ms
  • sms.confirmsent once
The problem

Off-the-shelf chatbots recite scripts, cannot see the real calendar and are confidently wrong. Voicemail and unanswered WhatsApp after six.

The approach

A prompt compiler builds the agent from onboarding answers; a server-side booking guard checks calendar, duration and rules; Flow Studio lets tenants draw routing. Booking as a guarded tool, not a sentence.

Architecture & AI decisions

Multi-tenant from the first line

Each business has its own configuration, hours, services, flows and channels, isolated at the data layer. Behaviour that must not vary — escalation and refusal rules — is fixed outside the tenant's reach.

01

Prompt compiler turns wizard answers into instructions

Prompt compiler turns wizard answers into instructions; tenants never write a prompt

02

Booking guard rejects anything that would overlap before it reaches th

Booking guard rejects anything that would overlap before it reaches the calendar

03

Flow Studio node graph for routing

Flow Studio node graph for routing: new vs existing, urgent vs routine, when to hand over

04

Test Lab runs scripted conversations per tenant on every change

Test Lab runs scripted conversations per tenant on every change; failures block deploys

What shipped.

01

Onboarding wizard

Compiles a working agent from answers
02

Tool-calling engine

Server-side booking guards
03

Flow Studio

Node-graph routing editor
04

Test Lab

Blocks bad prompts before deploy
05

Voice channel

Inbound telephony
06

WhatsApp & web widget

One script tag to install
07

Operations console

Transcripts, tool calls, cost
08

Multi-tenant core

Isolated configuration per business
Result
3channels per tenant
0overlapping bookings
1afternoon to configure
model choice — config, not rewrite
Next.jsTypeScriptPrismaPostgreSQLVercelOpenAI · Anthropic · GroqWhatsApp Business APITelephony / WebRTCNext.jsTypeScriptPrismaPostgreSQLVercelOpenAI · Anthropic · GroqWhatsApp Business APITelephony / WebRTC
Selected work

More in production.

All case studies
Retail · AI sales agent

A WhatsApp agent that answers from the catalog, quotes the right market's price and knows when a human should take over.

get_variants, check_stock, quote_price, lead_time against the Prisma catalog; retrieval pipelines evaluated and rejected for this size.

190variants answered from live data
8markets priced correctly
0vector databases
Read the case study
WhatsApp · sales agent
14
detect_market(phone)4ms
get_variants('halo floor')31ms
quote_price(H-220, AE)8ms
Engineering & compliance · SaaS

A cable calculator that turns two British Standards into a subscription product.

A permanent free plan converts on limits; Stripe subscriptions with plan gating, verified webhooks, proration and dunning; admin matches Stripe.

2British Standards encoded
1permanent free plan
15Next.js major version
Read the case study
Compliance calculator · Pro
16 mm²
Load 32 kW · 45 m · outdoorwithin limit
Subscriptionactive
Webhook invoice.paid

Have a system
like this in mind?

Send the problem, not a brief. An engineer replies within one business day.

Start a project