Ο Γιατρός της Πείνας

The menu had the answers. Customers still had the questions.

A traditional restaurant website is good at telling you what exists. It’s less useful when the customer asks: “What should I eat?” “How much food do we need for five people?” “What should I order for an event?”

Those are small questions individually. Repeated across customers, they become friction: more decisions for the customer and more repetitive conversations for the restaurant.

The existing website was primarily informational. The challenge wasn't simply to make it look newer. It was to turn it into something useful.

And whatever we built had to feel unmistakably like Ο Γιατρός της Πείνας — not like a generic AI chatbot dropped into a restaurant website.

We gave the website a Nurse.

We rebuilt the experience in Next.js, TypeScript and Tailwind CSS, then introduced an AI-powered layer around a very human problem: “I’m hungry. Help me decide.”

The result is «Ρώτα τη Νοσοκόμα».

Instead of an empty chat box asking “How can I help you?”, the Nurse starts from three real situations: “I don't know what I want. I just know I'm hungry.” “The whole group is hungry. Save us.” “We have people to feed. A lot of people.”

From there, the interaction changes with the situation. One hungry customer gets a recommendation based on appetite, preferences, dietary considerations and drinks. A group gets quantities adapted to the number of people eating. An event becomes a structured enquiry containing the information the restaurant actually needs: guests, date, budget and dietary requirements.

And when the questions are over? The Nurse passes the case to the Doctor. The recommendation arrives as a prescription, using real dishes, real prices and calculated quantities.

Not AI for the sake of AI. AI given a job.

AI handles judgement. Code handles truth.

This distinction shaped the entire system. The model is good at interpreting preferences and deciding what combination of dishes makes sense. It should not be trusted to remember whether something costs €3.60, whether two similarly named dishes are actually the same product, or whether five people need one or three portions.

So we separated intelligence from facts.

The restaurant menu became structured data. Each item has its own identity, category, price and role. Shareable dishes can carry serving information. Similar products from different categories remain explicitly different products.

The AI works inside those boundaries. It can reason about the customer’s answers and propose menu-item IDs. The application then validates those IDs against the real menu and uses deterministic TypeScript logic for quantities, prices and totals.

The model recommends. The system verifies.

We applied the same thinking to the experience itself. We didn't want the technology to dictate the interaction. We started with the restaurant's identity and designed the technology around it.

Less searching. Less guessing. Better questions reaching the restaurant.

The new experience turns common moments of uncertainty into guided interactions.

A customer can go from “I have no idea what I want” to a complete order recommendation without searching through every item on the menu.

A group can receive suggestions based on how many people are actually eating rather than guessing portions.

An event enquiry reaches the restaurant already structured around the information needed to continue the conversation.

For the customer, that means fewer decisions between hunger and food. For the restaurant, it creates the potential to reduce repetitive questions and receive better-qualified enquiries.

And for the brand, the website becomes more than a digital menu. It becomes another expression of the restaurant itself.

Specific problem. Specific system. Built around how the business actually works.

That’s the kind of technology we build at Deep Tail Tech.