We went to build a digital factory. We found the missing layer.

022022

Writing the glue ourselves

Yasir Tunçer, now our CEO, and Barış Velioğlu, now our CTO, were in that room as consultants. In practice their job became carrying messages: walking from one party to the next, explaining what the other side needed, convincing each one to do its part. It took an enormous amount of time, and most of the time it did not work.

So Barış stopped waiting. He started writing the missing parts himself. A script that read the machines and fed the MES. A small app that pulled orders from the ERP. Another that pushed quality data to a dashboard. One by one, the systems began to talk.

Every script worked, but together they were still not a system. And while we were writing them, the industry was putting a name to what we were doing by hand: the Unified Namespace, one shared structure where every system publishes what it knows and finds what it needs.

That was the lesson we took home. No single platform will ever solve every problem in a factory. So the communication between them is not a nice-to-have. It is the foundation everything else stands on.

032023

Version 1

When the digital factory was finished, the gap was still there. None of the players in that room had closed it. So we asked ourselves the obvious question: if nobody else will build this layer, why don't we?

Version 1 of MaestroHub was built for the biggest plants. It started as a distributed system of microservices, made to scale across large enterprises. And it worked. It went into real factories, at real customers, and solved the problem we had watched stall in that room.

But it asked a lot of the hardware. The more plants we saw, the clearer it became what we really wanted: one product that runs on a Raspberry Pi next to a machine and as a highly available cluster across sites. One product and one architecture, where the configuration decides whether it behaves like a single program or a distributed system.

Version 1 could not become that, so we needed something new.

04June 2025

The quiet rewrite

By 2025, AI could write code, and many of the best developers we knew did not trust it. They saw messy architecture, new bugs, and confident answers that were wrong.

Barış saw something else. An AI agent does what its foundation tells it. Give it enough context, a clean architecture, written principles and real domain knowledge, and it stops wandering. It follows your rules.

It also meant a hard decision. We had put years into version 1. So Barış did not say the whole plan out loud. He asked the team for two months to rewrite "a few modules" and see what AI could really do. Six weeks in, he had his answer. It was time to rewrite MaestroHub from scratch: modular, as light as possible, still able to scale, and free of every mistake version 1 had taught us.

For six months he built version 2 with AI agents, while the rest of the team, three or four people, kept version 1 running for our customers. Nobody else touched the new code. When the beta came together, it could do almost everything version 1 could, and building data pipelines was far more flexible. Only the connectors were behind. So the team joined: they built connectors, he built features.

If you really invest in the foundation, the AI stops going its own way. It follows the rules and principles you already built.

05February 2026

Live on stage

Eight months after the first line, on February 9, 2026, we showed version 2 for the first time, on stage at ProveIt! in the United States. It was close. The release was not done until the last day, and Barış was still writing features and fixing bugs on the flight over.

On stage we built a new connector, live, in front of the audience. To us it felt easy. To the room it was a surprise, and that surprise was exactly the point we came to make: with a strong foundation, real domain knowledge, and AI agents that respect both, a small team can build what used to take years.

People noticed something else they did not expect from industrial software. They enjoyed using it. You could pick your own avatar and your own background, and the screens felt calm and a little playful. That was on purpose.

Nobody says it, but everybody agrees industrial software should be boring. I wanted to build the first industrial application with some fun in it.

062026

Changing the job

After the launch we had to be honest with ourselves. We had a working product, tested from unit tests to full browser tests, and a pipeline engine of our own: one experience, instead of tools from different vendors stitched together and the visibility lost in the seams. But on paper, we did what other DataOps platforms did, and that was not enough for us. So we changed what the product does.

First, we made the product proactive. Most industrial software is reactive: something goes wrong, and you go hunting through the same pages you always see. MaestroHub tells a different story depending on what is happening. When everything is quiet, the screen is quiet. When something is emerging, the screen shows that, and the people it concerns are notified.

Then we did something that felt strange after all those months on the interface. We made the interface optional. Through MaestroHub's MCP server, people build and run their data operations from any MCP client, without opening MaestroHub at all, and always inside their own permissions.

That raised a question. AI can change a lot, very fast. Who changed what, and can you undo it? So we made governance granular. Every action lands in a tamper-evident audit trail. Everything you build has an owner you can hand to another person, every change is versioned so you can roll back to any point in time, and a pipeline can wait for a person to approve its next step.

07Now

Closing the loop

By then we had a strong DataOps platform and a good way to build a Unified Namespace. It still was not enough. AI could see the values, but not the factory: what a machine is, what it feeds, what it belongs to.

So we asked bigger questions. Why not a semantic layer, one language for the whole plant? Why not a knowledge graph, where every thing and every value knows how it relates to the rest? That is how AI understands context. That is how the loop between the factory and AI finally closes.

People stay in charge. AI can propose and act, and the system can stop it and ask the right person to approve or reject. Humans in the loop when it matters, and on the loop when it does not.

And there was one more group we kept meeting: domain experts who know exactly what they need to see, but are not software engineers. They should not have to think about deployment, security, who may open the app or who keeps it running. So they describe the app, AI builds it on MaestroHub's SDKs, and they deploy it and share it.

Today that is three products in one platform. The Context Engine to see the plant. The Knowledge Graph to understand it Coming soon. And App Studio to act on it Coming soon. Together, they are MaestroHub.