One Platform, Two Very Different Services
Ministerium is preparing public tests of two of its first vertical services from the end of Q3 2026: one for learning, one for catering. Their markets could hardly be more different. That is precisely the point.
Ministerium was conceived around a simple idea: digital services should adapt more naturally to people, instead of continuously asking people to adapt to new applications, interfaces and data structures.
The ongoing AI transformation has made that idea increasingly practical.
From the end of the third quarter of 2026, we plan to begin public testing of two Ministerium verticals that illustrate this approach from two very different directions.
The first is currently called Swiss Tuto, our implementation of Socratic Tutoring, initially focused on Switzerland.
The second still carries its development codename, Ctwo, and revisits a problem we first worked on more than twelve years ago: making self-ordering practical for the very large part of the catering industry that has never adopted it.
They share almost no business logic.
They share an architecture and a philosophy.
Swiss Tuto: AI should help students learn, not help them avoid learning
As the 2026–2027 school year gets underway in Switzerland, students, parents and teachers face a question that has become impossible to ignore.
Students already have access to extraordinarily capable general-purpose AI tools. But having access to AI is not the same thing as having a tutor.
In fact, used badly, AI can do exactly the opposite of what education requires: remove the intellectual effort from which learning comes.
Swiss Tuto starts from the opposite principle.
Its purpose is not to produce homework for students, bypass exercises or provide polished answers that students have not learned to construct themselves.
The objective of Socratic Tutoring is to help a student understand.
That means explaining fundamentals, asking questions, identifying where understanding breaks down, adapting explanations, providing exercises and accompanying progress over time.
Sometimes the right answer from a tutor is not an answer at all.
It is another question.
This is not merely our preference. The emerging evidence around generative AI in education increasingly distinguishes between general-purpose AI that performs tasks for students and purpose-built educational AI designed to preserve cognitive effort and support learning.
That distinction is fundamental to Swiss Tuto.
The ambition is to provide the qualities expected from serious private tutoring — availability, patience, continuity, contextual understanding and adaptation to the individual student — without transforming AI into an academic shortcut.
From an AI tool to an educational context
There is another gap we want to address.
The problem today is no longer a shortage of AI tools.
It is almost the opposite.
Students can access models, search engines, calculators, image generators, translation tools and specialized applications. What is often missing is the contextual layer that turns these capabilities into a coherent learning environment.
A student should not have to understand which AI model, application or technical format is appropriate every time a problem changes.
The service should understand enough of the educational context to help select and coordinate the appropriate tools.
This is particularly important in scientific subjects.
Mathematics, physics and other sciences cannot always be reduced to a succession of chat messages. Students need equations, diagrams, graphs, annotations, structured exercises and visual manipulation.
The smartphone remains the natural device for much of this generation, but a conversational entry point does not mean forcing the entire educational experience into a chat window.
Ministerium can begin with a conversation and progressively present the interface that the task actually requires.
The conversation becomes the doorway, not the constraint.
Swiss Tuto will initially concentrate on scientific disciplines, while progressively exploring languages, practical learning situations and selected subjects where the same tutoring principles make sense.
Ctwo: twelve years later, the problem is still there
The second public test has a very different origin.
In 2014, we developed a project called Carteplay with a small international team working between Singapore, Colombia and Silicon Valley. Carteplay Inc. was incorporated in Texas.
The idea was early: use the smartphone to let customers order directly in bars and restaurants.
We made a strategic mistake.
Instead of concentrating intensely on restaurant self-ordering and establishing the concept with major operators, we attempted to generalize the technology to on-site commerce.
At approximately the same period, companies such as McDonald's were beginning the large-scale deployment of self-ordering kiosks that would make this behaviour familiar to millions of consumers.
Carteplay ended before completing its first funding round following disagreements inside the team.
Twelve years later, the underlying problem has not disappeared.
So in 2026 we rebuilt the idea from Ministerium.
The codename is therefore Ctwo — Carteplay, a second time — although the service will receive a definitive name before commercial deployment.
But this is not a reconstruction of the 2014 product.
AI changes what is possible.
Self-ordering without installing a self-ordering system
Self-ordering works extremely well in many large chains.
But its economics and operational complexity remain much harder to justify for a huge number of smaller restaurants, bars, cafés, temporary venues and other catering situations.
Installing kiosks, POS extensions, payment terminals, kitchen-display systems and dedicated applications can require equipment, integration, configuration, training and support.
Ctwo begins with a different question:
What if a catering operator could obtain most of the benefits of self-ordering without first deploying a conventional self-ordering infrastructure?
Our target is deliberately demanding.
We want to determine whether a very small team could configure and begin using Ctwo within tens of minutes rather than days or weeks, using equipment they already possess.
One of the test scenarios is intentionally extreme: could two people operate a busy bar serving up to 200 seated customers per hour, assisted by Ctwo, without deploying dedicated kiosks, specialized ordering terminals or a conventional technology stack?
Could a new operator begin configuring the service and accept its first walk-in customer orders approximately one hour later?
These are test objectives, not claims of results already demonstrated at commercial scale.
That distinction matters.
The public tests are intended precisely to discover where the model works, where it fails, and what must change.
A customer should not need instructions before ordering a drink
The consumer side follows the same philosophy.
For most customers, self-ordering should require virtually no preparation.
There should be no application to discover, download, register and learn before being able to place an order.
For the more than three billion people already using WhatsApp, the initial interaction can begin somewhere they already know.
The objective is that within seconds a customer can ask questions, understand what is available, express preferences, configure an order and act on it almost as naturally as when interacting with an attentive professional waiter.
Sometimes this may happen in a conventional restaurant.
Sometimes it may happen in less conventional places where deploying a traditional ordering system would make little economic sense.
And again, conversation does not mean that everything must remain inside WhatsApp.
When structured selection, visual presentation, payment or another interaction is better handled through a web interface, Ministerium can move naturally into that interface while preserving the context of the conversation.
The service should adapt to the task.
The customer should not have to adapt to the software.
The same principles, in two completely different markets
At first sight, private tutoring and catering have very little in common.
From Ministerium's perspective, however, they expose many of the same problems.
A service needs to understand context.
It needs to know who is interacting with it, while asking for no more identity than necessary.
It must distinguish conversation from authoritative action.
It must know when an AI model is useful and when deterministic software should take control.
It must preserve state without forcing the user to maintain rigid forms.
It must move naturally between conversational and structured interfaces.
It must remain understandable to people who have never received training on the system.
And it must combine simplicity with security, privacy and integrity.
These principles also explain why we pay particular attention to identity, GDPR, permissions, copyright and the boundaries between generated information and authoritative data.
Our objective is not to put an LLM in front of an existing application.
It is to reconsider how the service itself can be delivered.
Why can we test two verticals at almost the same time?
A few years ago, attempting to build two serious products simultaneously for education and catering would have been irrational for a small development organization.
The domains are too different.
The interfaces are different.
The workflows are different.
The regulatory questions are different.
The users are different.
What has changed is not simply that AI writes code faster.
AI is progressively removing some of the rigidity that historically forced software developers to anticipate and encode every possible interaction into forms, menus, data structures and predefined sequences.
Natural language can now become an interpretation layer between human intention and structured software.
Multimodal AI can help interpret documents, images, speech and complex input.
Models can assist developers in understanding, testing and modifying increasingly large systems.
And specialized agents can accelerate parts of the development process that previously required substantially larger teams.
Ministerium has also been developing its own development methodology and internal tooling around these capabilities.
The consequence is not that software development has become effortless.
Quite the contrary: as AI makes it possible to build more, architecture, verification, security and discipline become even more important.
But it does mean that a small organization can now reuse a common service architecture while concentrating development effort on the genuinely different domain logic of each vertical.
That is what makes these two concurrent tests possible.
One architecture should survive very different realities
Launching Swiss Tuto and Ctwo close together is therefore not only a scheduling decision.
It is itself an experiment.
If Ministerium's architecture only works for tutoring, we have built a tutoring platform.
If it only works for restaurants, we have built a catering platform.
But if the same foundations can support a student learning mathematics and, almost simultaneously, a customer ordering from a busy bar — while each experience remains natural to its own environment — then we begin to validate something more general.
That is the hypothesis behind Ministerium.
Not one universal application.
Not one giant chatbot attempting to do everything.
A common service layer capable of supporting highly specialized verticals while allowing each one to behave according to its own users, rules and reality.
Public tests are for learning
We are deliberately calling what begins from the end of Q3 public testing, not a commercial launch.
There will be things to correct.
There will be assumptions that turn out to be wrong.
There will almost certainly be interactions that looked obvious during development and prove less obvious when used by students, families, teachers, restaurant operators and customers in real situations.
That is exactly why these tests matter.
AI makes it possible to build extraordinarily quickly.
It does not make reality optional.
For Swiss Tuto, the test will be whether AI can support learning without replacing the effort required to learn.
For Ctwo, it will be whether a sophisticated operational service can feel almost trivial to deploy and use.
For Ministerium, the test is larger:
Can very different services be brought to people where they already are, while making technology progressively less visible rather than asking users to become progressively more technical?
We will soon begin finding out.