Deployment walkthroughVoice AI & product development
Foyen: from voice demos to a front desk.
I started by building voice demos for local businesses. Foyen grew into a product an owner could set up, try and check from their phone.
In briefTL;DR · 2 min read
- Started with voice demos tailored to individual businesses.
- Built the setup, call playbook and mobile review experience around the voice engine.
- Connected Google Calendar and checked a real booking in a controlled product test.
Evidence checkpoint: March–June 2026 build · public demos checked September 10
Hear Foyen
Two scripted demos with synthetic voices, made in June 2026 to show the conversations we were designing.
A broken AC and a same-day request 0:53
Read the script
Foyen: Thanks for calling Summit Air, this is Anna. How can I help you today?
Caller: Hi, yeah, our AC just died and it is boiling in here. I've got two kids home today. Any chance someone can come out?
Foyen: Oh no. Okay, let's get you cooled down. Quick question so I send the right tech: is it running but blowing warm air, or completely dead?
Caller: It's running, but the air coming out is warm.
Foyen: Got it, that's usually fixable same day. I've got a technician finishing a job nearby. I can have him there between two and four this afternoon. Does that work?
Caller: Two to four works. That's amazing.
Foyen: Perfect, you're booked. I've got your number from caller ID. I'll text you a confirmation now, and again when he's twenty minutes out. Anything else?
Caller: No, that's it. Honestly, I expected to get voicemail.
Foyen: We answer every call! Help is on the way. Stay cool!
A plumbing call switches to Spanish 0:43
Read the script
Foyen: Thanks for calling Vega Plumbing, this is Anna. How can I help you?
Caller: Eh, hola... ¿habla español? Tengo una fuga de agua debajo del fregadero, ¡hay agua por todo el piso!
Foyen: ¡Claro que sí, le ayudo en español! Primero, cierre la llave de paso debajo del fregadero, ¿de acuerdo? ¿La fuga es constante?
Caller: Sí, es constante. Ya cerré la llave, pero necesito un plomero hoy.
Foyen: Entendido. Puedo enviarle un plomero hoy mismo, entre cuatro y seis de la tarde. ¿Le funciona?
Caller: Sí, entre cuatro y seis está perfecto. Muchas gracias.
Foyen: Perfecto, queda agendado. Le mando la confirmación por mensaje de texto. ¡Que tenga buen día!
Hear all four demos on Foyen ↗
The interface
The mobile web interface, using sample records. Names, call counts and payments shown here are demo data.
My part. Product direction, interface design and the AI-assisted build: business playbooks, onboarding, the voice connection, calendar booking and call review. I demonstrated the early agents to businesses and worked through the product with my co-founder.
AI assistance. AI coding tools helped implement, debug and test the software. Existing voice models power the conversations. I chose the workflows, directed the design and reviewed what we built.
- 1
Teach it the business
Services, questions and when to hand over
- 2
Take the call
Understand the request and use the connected tools
- 3
Show the owner
The outcome, details and anything still needing attention
It started with individual businesses.
Before Foyen, I was building and showing voice demos for local businesses. A restaurant and an employment-services office needed different answers, questions and handoffs. Each demo made me work through what that particular front desk actually did.
On a run in May, I noticed a mobile pet spa and thought about the person doing appointments all day, then still answering the phone after work. I wanted them to be able to teach an agent their business, try it and see what happened afterward.
That became Foyen. I kept the voice experiments in a separate lab and built the product around them: setup, a call playbook, an owner interface and records of what happened on each call.
The dashboard was a wall of text.
In a June walkthrough with my co-founder, the data was there, but the experience was too much work. I called it a wall of text. We also caught an onboarding mistake: choosing a med spa still showed furnace repair in the playbook.
I pushed the next version toward a few useful things on a phone: what needs you, what is booked and what happened on the call. The public demo above shows that design. The transcript and detailed records sit behind the summary, ready when you need them.
Technical lens
The product separates the voice bridge from the web interface. The bridge connects Twilio calls to Gemini Live and business tools; the web app reads account-scoped records through the same REST API used by the operator tools. The public interface demo uses fixed sample data.
Then I checked a real booking.
In the June 9 test, I connected my Google Calendar through Foyen. The product created a real event and showed the booking, the deposit link and their status in the call record.
That gave me something concrete to inspect beyond how the conversation sounded. The calendar event existed. The payment was still pending, and the customer text had not been sent.
Technical lens
The recorded test connected Calendar through Google OAuth, checked a booking window, created an event and read it back. The deposit was an operator-provided payment link. This was a controlled test on my account; paid customer traffic was not moved.
Evidence at a glance
| Check | Recorded result |
|---|---|
| Audio samples | Two scripted, synthetic-voice demos; four available on Foyen |
| Interface | Public mobile and desktop demos with sample records |
| June 9 calendar test | Real event created and read back on my connected calendar |
| June 9 payment and text | Payment pending; customer text created, not sent |
The audio and interface are demonstration assets. The calendar result comes from the separate June 9 controlled test. Paid adoption, collected deposits and a completed customer rollout are not established by this case.
What to take away
The part I kept coming back to was the owner: how they set it up, try it, and make sense of the calls afterward.
Discuss your workflow