Voice Pilot · Solo project · 2025 to 2026

Can I trust an AI with my patients?

Voice Pilot is an AI receptionist for clinics. It answers every call, books visits into the clinic calendar, and confirms by text. Every screen I designed exists so a clinic owner can answer that question with evidence.

Visit voicepilot.live

Founder, designer, and builder. I shipped the whole company myself: product, conversation design, brand, pricing, code, and go to market.

The homepage is the product. A visitor starts a real call in the browser within ten seconds of landing. You can interview the receptionist before reading a word of marketing.

9:04 PM · The clinic is closed

Who answers when I lock the door?

A patient calls after hours. Maya picks up on the first ring, checks the real calendar, offers two open times, and books the visit while the caller is still on the line. A confirmation text lands before the patient puts the phone down.

Call transcript. The agent explains the clinic is closed on Sundays, offers three Saturday times with Dr. Aidin Ghotbi, and handles a request for a female dentist before confirming. — The voice is designed like a front desk hire: one thought per turn, answers under three sentences, energy matched to the caller.

The second call

Will it treat my patients like regulars?

Returning callers are recognized by the number they call from. The greeting uses their name and picks up where the last conversation ended. The demo line turns this into its own pitch: it invites every caller to call back tomorrow and see if it remembers them.

Demo line transcript. Maya greets Callebe by name on a return call and asks which business he is calling from, picking up where the last call ended. — A real transcript from the demo line. The caller tried a fake name. The agent stayed warm, kept the recognized identity, and explained itself.

“Callebe, good to hear from you again. If you’re calling as Mike today, I can roll with that.”Maya, on the demo line, July 2026

7:58 AM · Next morning

What happened overnight?

The owner opens Voice Pilot with coffee in hand. Every caller has a profile, matched by phone number, with a plain language summary of each call. Notes written on a profile become context the agent reads on the next call, so the front desk and the AI share one memory.

Trust is built by showing the work. The owner never has to wonder what the AI said: every call carries its transcript and its summary.

The calendar

Where did this booking come from?

One calendar serves the whole clinic, and every booking carries its origin: a teal dot for phone calls, indigo for the website widget, gray for entries the staff added by hand. Provenance answers the quiet question behind every AI product: why does the system believe this?

Bookings week view. One calendar for the whole clinic with color coded bookings and a legend: booked on a phone call, booked through the widget, added here in the dashboard. — The legend is the design argument in miniature: color used only where it carries meaning.

Setup, once

What will it ask, and can I change it?

Owners describe what the agent should gather in plain language, and each field becomes a column in their data. Fields marked Confirmed live are read back out loud on every call. The agent itself is organized the way you would explain it to a new hire: Brain, Conversation, After the call, Deploy.

Collect field builder. A field name and type, a plain language description of what to capture, and a toggle for the agent to confirm the field out loud on every call. — Plain words in, structured data out. No prompt engineering exposed to the owner.

Agent configuration sidebar grouped as Brain, Conversation, After the call, Deploy, and Access. — The information architecture teaches the mental model of an AI agent: Brain, Conversation, After the call, Deploy.

What exists today

Every number below is real and current.

Live

Product in production at voicepilot.live, callable right now

5 surfaces

Marketing site, dashboard, web widget, phone line, billing

What building it proves

The skills underneath the screens, in the order this page showed them.

Conversation design

Greeting logic, turn pacing, memory moments, and recovery behavior written as a designed script, then tuned against real transcripts.

Trust surfaces

Transcripts on every call, provenance on every booking, confirmed live fields, and notes the agent visibly reads. Confidence is always earned on screen.

Information architecture

An agent organized as Brain, Conversation, After the call, and Deploy. The structure itself teaches owners how AI agents work.

Commercial design

Pricing model, plan structure, and a physical mail campaign designed as one funnel that ends in a phone call.

Production code

Next.js, Supabase, realtime voice, Stripe, and telephony, shipped and maintained by one person.