← Back to portfolio
Own project · Proof of concept for WhatsApp reservations

A WhatsApp assistant that takes reservations without interrupting service

The project

El Olivo is not a client or a real restaurant: it is the sample restaurant I built this proof of concept on. I wanted to offer reservation assistants to restaurants and, before selling it, I needed to check that the whole thing works: not a demo chat, but real reservations against a seating capacity.

I gave it what any neighborhood restaurant has — a short menu, two seatings for lunch and dinner, limited capacity — and built it in full on that data, from the WhatsApp conversation to the front-of-house dashboard.

The problem it solves

It is the problem of almost any restaurant that takes reservations by phone:

  • The phone rings right at peak time, and almost always for the same things: booking, changing the time or asking about opening hours. The person who picks up is the same person who is waiting tables.
  • Changes and cancellations get written on top of what was already there, so nobody knows for sure how many guests are coming that night.
  • Nobody calls to remind people of their reservation, and tables left empty by no-shows are written off as lost.
  • A lot of people would rather send a WhatsApp voice message than call.

What I built

An assistant that answers on WhatsApp at any hour

It replies on the restaurant's usual number. It gives opening hours, location, parking and the menu of the day, and answers in whatever language people write in. There are no option menus and no "press one": the customer writes however it comes out.

Complete reservations from start to finish

It checks whether there is room before promising anything, collects the date, time, party size and special requests, and returns a confirmation code. After that the customer can look up their reservation, change it or cancel it in the same chat. Canceling always asks for explicit confirmation, so nobody loses their table over a misunderstanding.

It understands voice messages

If the customer sends a voice note, it is transcribed and handled just like a written message. For a lot of people that is the difference between booking and not booking.

Automatic reminder the day before

Every morning the system reviews the reservations for the next 24 to 48 hours and sends a WhatsApp reminder. Each reservation is marked as notified, so nobody gets the message twice.

A dashboard for the front of house

A web screen with today's and tomorrow's reservations, the expected number of guests and the full list, with filters by date and status. From there you mark a reservation as completed or cancel it, and the screen updates on its own when a new reservation comes in. There is a "next ten" mode to see at a glance what is arriving next.

Under the hood: WhatsApp via Twilio · Claude · Whisper for voice messages · Supabase · deployed on Railway

The result

  • The assistant answers, books, changes and cancels over WhatsApp without anyone on the floor having to step in mid-service.
  • The reservation book keeps itself up to date: new bookings, changes and cancellations are recorded on the spot and show up on the front-of-house screen.
  • The day-before reminder goes out automatically and is marked reservation by reservation, so nobody gets the notice twice.
  • Voice messages are understood just like a written message.
  • It is built end to end, not as a mockup: adapting it to a real restaurant means changing the menu, the seatings and the capacity.
  • What it does not have yet is a real service behind it: it works in full, but it has not been used in a restaurant with real customers.

What I would need from your restaurant

  • One conversation to understand the menu, the seatings, the capacity and how reservations really work at your place.
  • The restaurant's WhatsApp number and little else: there is no need to change numbers or tell customers anything.
  • Zero technical staff. Nobody at the restaurant has to learn any new tool beyond looking at a screen.
  • Not one day of service stopped, no strange migrations, and no touching the register or the POS.

Do you have a similar case?

30 minutes by video call. If I don't see a clear opportunity, I'll tell you.

Let's talk for 30 min, free

More cases like this

See all projects
Own project · WhatsApp agent

WhatsApp Assistant for an Academy

Answers parents on WhatsApp at any hour: handles questions, recommends a course, walks them through payment and hands the conversation to a person when needed.

WhatsApp API · Claude API · Whisper
Own project · Personal assistant

Voice Task Assistant on Telegram

A voice note on Telegram becomes tasks sorted by project, which you can see and move on a board.

Telegram · Whisper · Claude API
Conversational agent · Voice

AI Voice Advisor

A voice agent available 24/7 to serve customers and generate opportunities.

ElevenLabs · React