Custom AI for the problems that off-the-shelf software has already failed to solve. Designed with your people, built to survive contact with your actual business, and licensed so that you are never stuck with us.
Somewhere in your business there is a process that is expensive, repetitive, and stubbornly immune to the software you already pay for. It probably involves a spreadsheet that three people are afraid to sort. Off-the-shelf does not fit it, because if it did you would have bought it years ago and we would never have met. That is our work: the systems that have to be made, because they cannot be purchased.
We start from your business and work toward the technology, never the other way round. That sounds obvious. It is also the single thing most AI projects get backwards, which is why so many of them end their lives as an impressive demo that nobody uses. We would rather build you something dull that runs every morning at six than something brilliant that only ever ran once, in a meeting.
Anyone can build a demo. A demo runs once, on clean data, in front of people who want it to work. Production runs on a wet Tuesday, on whatever your suppliers actually sent you, in front of nobody.
The test of a build partner is not the demo. It is what happens on the day you decide you would rather work with somebody else.
Built for companies with a real process, real data and a real problem, who have heard quite enough about how AI will transform everything and would now like one specific thing to work.
The industry has started calling this forward-deployed engineering. We called it doing the job for eight years before it had a name, and we are not going to charge you extra for the vocabulary.
Five things we have done repeatedly and therefore know the true cost of. If your problem is not on this list, tell us anyway. The useful answer might be that you do not need us.
Letters, scans, PDFs, emails, and the fax that somehow still arrives. Ingested, read, understood, and turned into structured data your systems can actually use. This is the one we have done at the largest scale, and the one that most reliably pays for itself.
Asking your business a question in a sentence and getting an answer grounded in your real records, rather than a confident invention. Our AI Agent for CRM is exactly this idea, productised. If your shape is different, we build to your shape.
The unglamorous one, and usually the valuable one. Not optimising a step, but wiring the whole process together so that a case travels from start to finish without anyone re-typing anything into a fourth system.
The institutional memory currently held by the colleague who is about to retire, or buried in four thousand PDFs nobody can search. Made findable, grounded in the source, with a citation you can go and check.
Systems that take action, not just answer questions. Every action logged, and the consequential ones held for a human to approve. We design the brakes before we build the engine. It is an unfashionable order and it is the correct one.
Four stages. The first one is free and the second one is deliberately small, because you should be able to find out whether we are any good before you have spent real money.
You tell us what is going on. A rough sketch is plenty. We ask a great many questions about your business and comparatively few about technology, and at the end we tell you honestly whether there is something here worth building and whether we are the right people to build it. Sometimes the answer is no on both counts. You will get that answer straight, on the call, for free.
Then we come and look at the real thing. The real documents in their real formats, the real workflow, and the people who do the work and know exactly where it hurts. Not a workshop about the future of your industry. You end up with scope, architecture, a plan and a number you can take to your board. It is priced per engagement and kept small on purpose, because its job is to make the big decision safe, not to be the project.
We build with your people rather than at them, in short cycles you can watch. Then it goes into a pilot on genuine data with genuine users, which is where you find out what is actually true. Things surface here that no discovery could have predicted, and that is precisely the point of meeting them in month two rather than month nine.
Live, monitored, and supported by the same people who built it, for as long as you want us around. Or handed over cleanly to your own team, with the licence and the documentation that let them run with it. Both endings are genuinely fine by us. Only one of them is fine by most suppliers, which tells you most of what you need to know about most suppliers.
Priced per engagement. There is no public price list, because there is no standard build; what it costs depends on what you are actually trying to do and what state the data is in. The first conversation costs nothing, and if we are not the right fit we will tell you during it, rather than three invoices later.
Here is the part most suppliers leave in a PDF you receive after signing. We keep ownership of the code we write. You get a licence to it that is perpetual, unlimited and non-exclusive: run it, change it, extend it, host it wherever you like, and hand it to another development team to work on. It never expires, and it never costs you anything further.
The one thing that licence does not let you do is take our work, turn it into a product, and sell it against us. That is the whole restriction. Now read the rest again, because what matters is what is missing: there is no renewal you must sign to keep using what you already paid for, no clause that makes leaving expensive, and no hostage situation dressed up as a partnership.
You could fire us on a Friday and still be running on Monday, on your own hosting, with your own team. We would be sad about it. You would be fine. That is the deal, and we think it is the only honest one to offer.
Hostable in Europe, and on your own infrastructure if that is where you want it. We do not train anything on your data. Sovereignty is not a position we adopted when it became fashionable in Brussels; it is how we have built since before anyone was asking us about it.
Every automated action and every human one, traceable after the fact. Partly because that is what regulated clients require of us, and partly because it is the only way anyone can honestly tell you what a system did last March.
Built from open, swappable parts, with the model layer kept replaceable on purpose. This field reinvents itself every few months. Anything welded to today's best model is tomorrow's migration project, and you will be the one paying for it.
Not a well-known name on the contract and a stranger on the calls. These are the people who will be in your standups.
Architects the solution and stays close to the build. Educated in construction engineering, applied psychology and business, then twenty years advising boards and governments before he stopped advising on other people's technology and started building his own. An odd combination that turns out to be the right one, because the job is mostly understanding a business well enough to know what is worth automating.
Based in Leeds. Leads the team that turns a design into running software, and taught developers to build on our platform at two university hackathons in 2019. He is the one who tells you honestly how long it will take, including the parts you did not think to ask about.
Electrical engineering graduate from the University of Vienna. Builds the conversational agents and the operational guardrails that make them safe to point at real business data. He is the reason the brakes exist before the engine does.
Based in Novi Sad. Works the commercial side: the business case, the numbers, and the awkward question of what this will really cost and really return. He asks it early, while it is still cheap to hear the answer.
Behind the four of us sit around a hundred specialists we have worked with for years, and development teams across Croatia, Serbia and Montenegro. Practically, that means we can staff a build properly without sending you back out to find another vendor, and it means we can afford to say that one is outside our expertise, but we know exactly who to ask.
We are a member of AI Factory Austria's AI:AT Hub, part of the €2.6 billion European AI Factory initiative. We hold cooperation agreements with universities in Austria, Serbia and Montenegro, and we have organised six international hackathons, including with ÖBB, Deutsche Bahn and Infineon. That is not a badge collection. It is how we stay current in a field that rewrites itself every few months, and a good deal of what we know about where this is heading comes from being in those rooms.
The best builds we have done are the ones where, by the end, your people and our people were indistinguishable at standup. That is not a nice-to-have. It is where the knowledge actually transfers, and it is the part of the work we enjoy most.
The clearest way to explain what we do is to show you one we finished. A client managing parking across 150+ locations, owned by different parties, under rules that are legally enforceable, was chasing violation fees through a process held together by spreadsheets, email, letters and the institutional knowledge of people who were quietly losing the will to live. To invoice a fee you must first identify the vehicle holder, and that means writing to dozens of authorities across jurisdictions, each replying in its own format, on its own timeline, sometimes in weeks, sometimes in months, sometimes not at all.
Nobody sells software for that. So we built it: one platform covering the whole lifecycle, from the moment the camera sees the car to the moment the money arrives. AI reads the authority replies and matches them to the right case. The system prepares validated SEPA payment files. Every action, automated or human, is logged. It has been running in production for years, and it works well enough that we now sell it as a product in its own right, which is the most honest thing we can tell you about the quality of a build.
| The parking fee process, before and after the system we built. | Before | After |
|---|---|---|
| Data | Scattered across spreadsheets, email, legal software and paper files | One platform. One source of truth. |
| Reading the post | A person opens each authority letter and types the data out by hand | AI extracts it in seconds, and flags the cases where it is unsure |
| Case matching | Manual lookup and cross-referencing across systems | Matched automatically, with suggestions where it is genuinely ambiguous |
| Money | Spreadsheet calculations and manual bank transfers | Validated SEPA exports, grouped by authority |
| Traceability | No audit trail. Things fall through the cracks. | Full history on every case. Every action logged, automatic or human. |
| Accuracy | 80 to 90%, with the errors found late and expensively | 99%+, with validation checks and alerts |
| Turnaround | Days to weeks per case | Minutes, from response to fee collection letter |
| Scale | More cases means more staff, doing work that makes them leave | Ten times the volume with the same team, doing better work |
Parking is one example. Here is some of what else we've shipped, plus the platform behind two of them: 0bsnetwork, one of the few blockchain platforms built for real business use rather than cryptocurrency speculation. Instead of smart contracts (a name that is neither smart nor a contract) it uses data transactions: business data stored as plain, readable key-value pairs, understandable without a computer science degree.
The world's first platform for tokenizing and crowd-investing in real-world physical assets, built for Eloop's fleet of 200 Tesla Model 3 and Model Y cars in Vienna.
The world's first euro-backed stable token issued by a bank on a public blockchain, built and run for Raiffeisen Bank International.
A platform for fractional investing in real estate.
A trading platform for physical agricultural commodities, connecting banks, quality-control companies, insurers, warehouses, logistics and transport in one process.
A real estate platform.
A platform built for the Croatian Ministry of Health to track and prevent medicine shortages.
Discovery runs in weeks. A first useful thing in production is months, not years, and not fortnights. If somebody promises you a production AI system in two weeks, they are describing a demo and hoping you will not notice the difference until the invoice has cleared.
No. Arriving with a fully specified brief is rare and, frankly, a bit of a warning sign, because it usually means somebody decided on the solution before they finished looking at the problem. Come with the problem. Working out what to build is the job, and it is the part we are best at.
Yes, and we are pleased when they do. The licence allows it, the code is documented for it, and the projects that go best are the ones where your team gradually takes it over. That is not us losing work. That is the work having gone well.
Very often, yes, and where that is true we will say so on the first call. Buy the standard thing for the standard problem. Build only where the process is genuinely yours, which is usually the place your actual advantage lives. Anyone who tells you to build everything is selling build hours.
Sometimes. We will look at it properly and tell you whether it is worth saving or whether the kindest and cheapest thing is to stop. We have said both. Neither conversation is fun, and the second one is usually the more valuable.
Whichever is genuinely right for the job, and we keep that layer replaceable so the answer can change without a rebuild. Sovereign European models where that matters to you, and we will explain the trade-off in plain terms rather than hiding it. Choosing a model is an engineering decision, not an identity.
A rough sketch is plenty. You do not need a brief, a budget, or the vocabulary. Tell us what the problem actually is and we will tell you whether it is worth building.