Case study 01 · Voice AI
VeraVoice
An AI receptionist that answers every call on the first ring, books the appointment and screens out spam, on a voice pipeline we own instead of one we rent by the minute.
In production, not in a deck.
Figures from VeraVoice production deploymentscalls a month handled in production, day and night.
of caller intents understood correctly on live calls.
voice round trip. Under the point where people notice software.
per minute. We own the voice pipeline, so there is no middleman.
The most expensive callis the one nobody sees.
It comes at 7pm, or on a busy Saturday, or while the one person who answers phones is with a patient. It rings out. The caller doesn't leave a voicemail, because leaving a voicemail feels like effort and dialling the next result on Google doesn't.
Nobody records that loss. It shows up later as a quiet week.
For a small practice, both obvious fixes are bad. A night receptionist costs more than the calls are worth. A voice-AI platform charges per minute forever, on top of model and telephony costs, so the bill grows exactly as fast as the business does.
9:14 PM. The front desk is closed. The phone isn't.
Three ways to answer the phone, and why we didn't pick any of them.
Hire an evening receptionist
Covers a few extra hours, not 24/7. Costs a salary to catch calls that arrive in bursts.
Rent a voice-AI platform
Fast to demo. Then every minute of every call carries a platform fee on top of the model and the phone line.
Voicemail + IVR menu
Cheapest option. Also the one callers hang up on. “Press 2 for appointments” is where bookings go to die.
So we built the voice pipeline ourselves. The clinic owns the agent. Nothing sits between the business and its own phone calls.
Don't take our word.Hear it.
A new patient calls after hours with a chipped tooth. Press play and watch Vera check the calendar, book the slot and text the confirmation, live.
Press play. Actions appear here as Vera takes them.
Demo call for illustration, recreated with synthetic voices. Clinic, caller and doctor names are fictional. Real client calls are never published.
A receptionist,and the consolebehind her.
01Live console
Every call as it happens: the transcript, the slot Vera offers and books, and the spam she blocked.
02Clinic dashboard
Calls, after-hours answers, bookings and bot health at a glance. Shown with demo data.
03Spam screening
Robocalls flagged with a confidence score and blocked, so the front desk stops ignoring the phone.
04AI bookings
The week fills itself. Every appointment Vera made, next to the ones the front desk took.
Owned,not rented.
Hover any block. The same agent runs on OpenAI Realtime and on Gemini Live, so “no vendor lock-in” is a config switch, not a slogan.
NESTJS · PYTHON
Vera core. Our own orchestration: persona, guardrails, spam classifier and tool calls. No per-minute platform in the middle.
Response time
Under three seconds round trip, the point where callers stop noticing they are talking to software.
What we own
The pipeline, the prompts, the classifier and the console. The client pays for models and telephony, not for a platform markup.
What are your missedcalls costing?
Estimate from your inputs. Move the sliders to match your own numbers.
Discovery
We listen to real call patterns: when they peak, what people ask, what a booking needs.
Voice design
Persona, tone, language and guardrails tuned per business, then tested on scripted calls.
Integration
Calendar, SMS, CRM and the clinic's existing number. No new phone system.
Live + tuned
Calls monitored in the console. Misses become fixes within days, not quarters.
“The team demonstrated a deep understanding of both recruiting and AI, seamlessly integrating cutting-edge technology with real-world interview scenarios. Delivery was lightning-fast.”
Your phone,answered tonight.
Common questions
What is VeraVoice?
An AI phone receptionist that answers every call on the first ring, day and night, books appointments, texts confirmations and screens out spam calls.
Who is it for?
Clinics, dental practices, med spas and home-service businesses that lose calls after hours or while staff are busy.
What is it built with?
NestJS and Python for the core, Twilio for calls and SMS, a Next.js console, and OpenAI Realtime and Gemini Live for the voice.
How long does a deployment take?
Typically 4 to 8 weeks to production.
More case studies
All case studies →PreOp Screen
Getting patients ready for surgery, without the phone calls.
View case study →Retail Insights
From last week's spreadsheets to dashboards that are fresh every morning.
View case study →Syphon AI
Reads 200-page tenders, flags the deal-breakers, and cites every answer.
View case study →Esperwise
Interviews the patient, listens to the visit, and writes the note the way the doctor would.
View case study →Recruiting NOW
An AI that interviews every applicant in German, at any hour, and explains every score.
View case study →Contract Blitz
Explains every clause in plain English, flags what bites, and tells you what to ask for.
View case study →Brands10x
An AI CEO, CFO, CMO, COO and Chief of Staff that plan together around the same goals.
View case study →WorkApply
Finds the jobs that fit, tailors every application, and applies for you.
View case study →BuildStudio
The design team trailer shops hire on Upwork, as one AI studio.
View case study →Your project, next.
Tell us what you want to build. You get a fixed, line-item quote after one call.
Start a project →