Roman Michalek found a gap between translation and the law, and built a method to close it

The problem found Roman Michalek before he had a name for it. Two years ago he began experimenting seriously with AI, and kept noticing the same thing: translation quality was being judged by the wrong yardstick. Everyone asked whether a sentence read naturally. Almost no one asked whether it still carried the same legal obligation it started with.

That gap is not theoretical. He points to a case in which the Dubai Court of Cassation reversed a ruling over a mistranslated Arabic arbitration clause. “Most people assume this is a translation-quality problem,” he says, “something better translators or better AI will eventually solve on their own.” He believes they are measuring the wrong thing.

A career of reading behaviour

Before any of this, Michalek spent years selling gold and real estate. He describes watching clients sign in good faith, then get hurt when meaning and obligation quietly came apart. He also describes an unusual habit he once thought was just personal: never seeing a word as only a word, always as a behaviour pattern. That habit, it turned out, was the method.

I didn't start this because I saw a market gap to exploit. I started it because I couldn't unsee the harm.

Refusing the easy word

The decision that changed everything, he says, was refusing to call what he does “translation.” As long as he used that word, people filed him next to agencies competing on price. He was neither a translation agency nor a lawyer, and for a long stretch nobody had a category for the work, even as people with real expertise clearly recognised its value. So he built the category himself: a layer that sits between translation and legal review and protects the obligation when a document crosses a language line.

In practice, the method treats a contract less like prose to be rendered and more like a set of promises to be preserved. Where a translator asks whether the words match, Michalek asks whether the duty, the deadline, the liability and the remedy still bind exactly as they did in the original. It is closer to an audit than a rewrite, and it is the part that ordinary translation was never designed to catch.

The early version was not a company at all. It was just him, running the check by hand on freelance work he was already being paid for. The first real sign came when a translation company that had been in business for fifty years paid him for it.

Almost everyone around you is going to discourage you, and that's not a sign you're wrong. It's just what building something new looks like from the inside.

What comes next

Today, he says, the work has moved past proof-of-concept, with completed and paid engagements across the UAE, Europe, and Japan. The challenge now is scaling judgment-heavy, one-person work without diluting it. His stated ambition is large, around $100 million in five years, but he frames success differently: that it actually protects companies the way they were protected the moment they first shook hands.

Ask what keeps him going and the answer is not the number. His mother passed away around the time he started building this, and in many ways it has become a tribute to her. The discouragement, the missing category, the long stretch of doing the work by hand before anyone had a word for it: he treats all of it as the ordinary cost of building something that did not exist before.

This profile is drawn from Roman Michalek's own account and reflects an early-stage venture; figures and engagements are as described by the founder and pending independent verification.