# Reachy PizzAIolo: When a Restaurant Sign Starts a Conversation

> Source: <https://dev.to/danuw/reachy-pizzaiolo-when-a-restaurant-sign-starts-a-conversation-4agf>
> Published: 2026-08-24 23:21:02+00:00

*Building an AI-powered conversational restaurant ordering experience with Reachy Mini and Seeed Studio's reTerminal E1003.*

What if a restaurant sign didn't just display the menu?

What if it could **talk to you, understand what you wanted, help you make a choice, update your order as you spoke - and show everything visually at the same time?**

That is the idea behind **Reachy PizzAIolo**.

It combines the personality and conversational abilities of [ Reachy Mini Lite](https://www.seeedstudio.com/Reachy-Mini-Lite-Kit-p-6702.html?sensecap_affiliate=bWJooAR&referring_service=link), the physical presence of the

**a sign you can actually have a conversation with.**

I built Reachy PizzAIolo as my entry for Seeed Studio's ** Interactive Signage Contest 2026**.

This year's contest focuses specifically on **restaurants and dining**, asking makers to rethink what interactive signage could become when combined with technologies such as AI, voice interaction, sensing and IoT.

That immediately made me wonder:

Why should the interaction start with the screen at all?

Touchscreens are already everywhere in restaurants. They work well, but increasingly they feel like another computer interface placed between the customer and the restaurant.

A conversation is different.

You can say:

"I'd like a Margherita and a Coke."

Then:

"Actually, make that two pizzas."

Or:

"What vegetarian pizzas do you have?"

Or simply:

"That's everything."

So rather than building another digital menu, I wanted to see what would happen if **conversation became the primary interface** and the display became its visual companion.

And of course, if the robot taking your pizza order happens to look like Reachy Mini, there was only one sensible name:

**Reachy PizzAIolo.** 🤌

The core interaction is deliberately simple.

A customer speaks naturally to Reachy.

Reachy understands the conversation, maintains an order, and communicates changes to the ordering service.

The E1003 updates to give the customer a persistent visual representation of what has been ordered.

Meanwhile, restaurant staff - whether in the kitchen or front of house - can see the order appear on the active-order dashboard.

The result is an interaction that combines:

**conversation + physical character + visual confirmation + structured restaurant data**

rather than relying on any one interface alone.

One of the most enjoyable milestones for the project was getting to use **Reachy PizzAIolo at an actual pizza party**.

Rather than only testing the interaction as a controlled demo, Reachy was able to take orders from real guests in exactly the sort of social setting the project was designed for.

It was a simple but useful test of the idea: people could approach Reachy, place their order conversationally, and see that interaction turn into a structured order.

And, perhaps more importantly, it was fun. A robot taking pizza orders naturally became part of the experience rather than just another ordering terminal.

*Reachy PizzAIolo on duty at a real pizza party.*

At the front of the experience is the [ Reachy Mini Lite](https://www.seeedstudio.com/Reachy-Mini-Lite-Kit-p-6702.html?sensecap_affiliate=bWJooAR&referring_service=link), the small expressive robot from Pollen Robotics.

Reachy provides something a touchscreen cannot: **physical presence**.

Its head movement, antennas, voice and reactions make even a relatively simple ordering interaction feel very different from navigating menus on a kiosk.

But Reachy isn't merely there as an animated mascot.

The conversation is connected directly to the ordering system.

When the customer adds, removes or changes something, Reachy isn't just generating a response - it is affecting the state of a real order.

Beside Reachy is the [ Seeed Studio reTerminal E1003](https://www.seeedstudio.com/reTerminal-E1003-p-6731.html?sensecap_affiliate=bWJooAR&referring_service=link).

The E1003 is particularly interesting for this project because it combines a large monochrome ePaper display with touch capability.

💡

Seeed Studio discount:If you're interested in the E1003, E1004 or Sticky, you can use codefor an additional discount where eligible.`CFST3GPA`

In Reachy PizzAIolo, its main job today is **live order review**.

As the conversation progresses, the display reflects the current state of the basket.

That matters because voice interfaces have a weakness: **spoken information disappears**.

If Reachy says:

"I've added two Margheritas, one sparkling water and a tiramisu..."

the customer shouldn't have to remember all of that.

The E1003 turns the conversation into something persistent and glanceable.

You can talk to Reachy while still being able to look at the order.

This also creates the basis for something more interesting later: a genuinely **multimodal ordering experience**, where speaking and touching the display become interchangeable ways of modifying the same order.

At a high level, the architecture is intentionally straightforward:

```
       Customer
          │
          ▼
    🤖 Reachy Mini
          │
          │ Conversation
          ▼
   🧠 Ordering Service / AI
        │             │
        │             │
        ▼             ▼
 🖥️ reTerminal     🍽️ Active
     E1003             Orders
```

The **ordering service** is the important layer in the middle.

Rather than treating the conversation as a stream of unstructured text, it maintains a structured representation of the customer's order.

That means the same order state can be used by several different interfaces.

Reachy can talk about it.

The E1003 can display it.

The restaurant dashboard can act on it.

And future interfaces could modify exactly the same underlying order.

Imagine walking up to Reachy:

**Reachy:**

"Hi! Welcome. What can I get for you?"

**Customer:**

"Can I have two Margheritas and a Coke?"

The ordering service extracts the items and quantities and updates the active basket.

The E1003 now shows:

```
YOUR ORDER

2 × Margherita
1 × Coke
```

Then the customer says:

**Customer:**

"Actually, change one of the Margheritas to a Diavola."

Reachy updates the order rather than starting again.

```
YOUR ORDER

1 × Margherita
1 × Diavola
1 × Coke
```

And when the customer is ready:

**Customer:**

"That's everything."

Reachy can recap the order and move it into the restaurant's active-order workflow.

That ability to **modify an existing structured state through natural conversation** is one of the parts I find most interesting.

The customer-facing interaction is only half of the system.

Confirmed orders are also made available to an **active-orders dashboard**.

That turns Reachy PizzAIolo from a robot demonstration into the beginning of a restaurant workflow.

The important distinction is that the AI isn't merely producing a conversation.

It produces **structured actions that another system can consume**.

That opens up obvious future integrations with:

The robot is simply one interface into that system.

During this project, I increasingly found that the interesting part wasn't really the robot *or* the screen.

It was the relationship between them.

It is very natural to say:

"Give me another one of those."

or:

"What do you recommend?"

or:

"No onions on that one."

A screen is much better at answering:

What have I ordered?

How many?

How much?

What choices are available?

So instead of asking whether **voice or touch** is the better interface, Reachy PizzAIolo explores what happens when each does the job it is naturally good at.

Conversation handles intent.

The sign provides context and confirmation.

The current prototype implements the core end-to-end experience:

A customer can build and modify an order by speaking naturally with Reachy.

The customer-facing E1003 reflects the current order as the conversation progresses.

Orders flow through to a separate active-orders view for the restaurant.

Most importantly, these aren't separate demonstrations.

They are different views onto the **same order state**.

There are several directions I would like to take this.

At the moment, conversational turn-taking is more rigid than I ultimately want.

The next step is allowing the customer to interrupt Reachy naturally:

"Actually..."

and immediately correct or extend an order without waiting for the robot to finish speaking, while the system keeps track of the conversation so far.

That would make the interaction feel significantly closer to talking to a human member of staff.

The E1003 is touch-enabled, which creates a particularly interesting next step.

Imagine Reachy saying:

"We have four desserts - I've put them on the screen."

The customer could then either say:

"I'll have the tiramisu."

**or simply tap it.**

Both interactions would manipulate the same underlying order.

Conversation when conversation is easiest.

Touch when touch is easier.

No separate modes.

Restaurant conversations are deceptively complicated.

An ordering assistant needs to distinguish between actions such as:

I want to continue moving more of this behaviour into explicit, structured **skills/tools**, leaving the language model to understand the conversation while deterministic services handle the actual restaurant operations.

That should make the system both more capable and more reliable.

Another direction I am particularly interested in is moving more of the AI stack to the edge.

A future version could use something such as Seeed Studio's [ reComputer Robotics J5012](https://www.seeedstudio.com/reComputer-Robotics-J5012-with-GMSL-extension-board-p-6682.html?sensecap_affiliate=bWJooAR&referring_service=link) as a local AI server for the restaurant.

**To be clear: the J5012 is not part of the current implementation.**

What interests me here is not necessarily putting a powerful AI computer behind every robot or display. Instead, a restaurant could potentially have **one local AI machine serving the whole location**.

That shared edge server could provide services such as speech recognition, text-to-speech, language models and AI orchestration to multiple Reachy or signage endpoints around the restaurant.

The customer-facing devices could remain relatively lightweight, while the more computationally intensive AI runs centrally within the store.

For a real deployment, that could offer an interesting balance between local processing and cloud services - potentially improving latency, resilience and privacy without requiring dedicated AI hardware at every ordering point.

Despite the name, there is nothing fundamentally pizza-specific about the architecture.

Reachy PizzAIolo is really an experiment in **conversational interactive signage for restaurants**.

The same pattern could work in:

And Reachy doesn't necessarily have to be the final physical interface either.

The broader idea is a service where **conversation, signage and structured business workflows all share the same state**.

The pizzaiolo is simply a much more entertaining way to demonstrate it. 🍕

I am making the project available so that the individual components and architecture can be explored and reproduced.

👉 **GitHub:**

[https://github.com/danuw/reachy-pizzaiolo](https://github.com/danuw/reachy-pizzaiolo)

The repository contains the application code and will continue to receive documentation and setup instructions as I tidy the contest prototype into something easier for others to reproduce.

🤖 **Reachy Mini Lite - the version used in this project**

[Seeed Studio product page – affiliate link](https://www.seeedstudio.com/Reachy-Mini-Lite-Kit-p-6702.html?sensecap_affiliate=bWJooAR&referring_service=link)

📖 **Reachy Mini source and documentation**

[https://github.com/pollen-robotics/reachy_mini](https://github.com/pollen-robotics/reachy_mini)

🤖 **Other embodied AI robots from Seeed Studio**

[Seeed Studio collection – affiliate link](https://www.seeedstudio.com/hugging-face-embodied-ai-robots?sensecap_affiliate=bWJooAR&referring_service=link)

🖥️ **Seeed Studio reTerminal E1003**

[Product page – affiliate link](https://www.seeedstudio.com/reTerminal-E1003-p-6731.html?sensecap_affiliate=bWJooAR&referring_service=link)

🏷️ **Seeed Studio discount code**

Use ** CFST3GPA** for an additional discount on the E1003, E1004 or Sticky where eligible.

📖 **reTerminal E1003 documentation**

[https://wiki.seeedstudio.com/getting_started_with_reterminal_e1003/](https://wiki.seeedstudio.com/getting_started_with_reterminal_e1003/)

🏆 **Seeed Studio Interactive Signage Contest 2026**

[https://www.seeedstudio.com/make-a-sign-interactive-signage-contest-2026](https://www.seeedstudio.com/make-a-sign-interactive-signage-contest-2026)

🎥 **Project video**

[https://youtu.be/t1Ayfm52QaI](https://youtu.be/t1Ayfm52QaI)

I started this project thinking about how to make a restaurant sign interactive.

I ended up asking a slightly different question:

What if the sign stopped behaving like a screen and became part of the conversation?

Reachy PizzAIolo is still a prototype, but it has already convinced me that conversational AI becomes much more interesting when it escapes the browser.

Giving an AI a physical presence changes the interaction.

Giving the conversation a persistent visual surface makes it more usable.

And connecting both of those to a real structured workflow is what starts turning the demonstration into a product.

There is plenty more I want to explore - particularly full-duplex speech, touch + voice interaction, structured skills and local AI.

But for now...

**Buon appetito! 🤌🍕🤖**

*This project was created as an entry for the Seeed Studio Interactive Signage Contest 2026. Some links in this article to Seeed Studio products are affiliate links. If you purchase through one of these links, I may receive a small commission at no additional cost to you. Where eligible, you may also be able to use discount code **`CFST3GPA`

** on the E1003, E1004 and Sticky. This does not affect the opinions or technical information presented in this article.*

**Tags:** `#ai`

`#robotics`

`#iot`

`#opensource`