
They Uploaded Thousands of Pages of Law. The Engine Didn't Understand a Word of It.
Founder Stories by Built in Baltics. A conversation with Taavi Vahter, co-founder of Juvo Legal.
In this article
A month of work, thousands of pages of legal documents, theses, and court rulings fed into the engine and it couldn't parse undefined legal vocabulary. The suggestions came out subtly, dangerously wrong. For a brief moment the team thought the whole thing was dead. Then they did the only sensible thing left: they started teaching their agent law, from the ground up.
How It Started
Juvo Legal exists because of a rejection.
One of the founders was pitching an AI solution to a financial institution. They passed on the solution but on the way out, they said something that turned out to be worth more than a yes: if a platform for legal compliance existed, we'd be very interested.
Around the same time, Taavi and the third co-founder had both decided to pursue law at university. Suddenly a legal-compliance platform wasn't just a market gap someone had pointed at but it was one they were actively becoming qualified to fill. They entered a hackathon with no real idea how it would go, what it would look like, or how it would work. It went well enough that they were accepted into a summer residency programme to start building it properly.

Building From the Baltics
Juvo Legal is, by design, deeply regional. The product specialises in Estonian and EU law, which means being from Estonia isn't incidental to what they've built because it is what they've built. The regulatory specificity that would be a limitation for a generic tool is the entire moat here.
The second Baltic advantage is cultural. Taavi points to a startup scene that's genuinely welcoming and thrives on creativity and new solutions and to the infrastructure that makes it real. Without places like ruum in Estonia, he says, a lot of startups would have nowhere to start from. Juvo Legal included.
How It Works
The platform is simple on the surface. A client uploads their documents; the engine analyses them, combs through thousands of laws and court rulings, and flags anything that isn't in order with the law as it currently stands.
It's built for organisations drowning in legal documentation and long, thorough terms of service, like financial institutions, telecoms, law firms. For them, the compliance phase shortens drastically.
Under the hood: the real asset is a database of over 66,000 Estonian and EU legal acts. The agent works as a citation verifier, checking every section against the actual statute text rather than trusting the model to remember it. In a domain where a confident hallucination is a liability, that verification layer is the product.
The signal that it was working came from the field itself. Real lawyers started reaching out asking how it was going, when it would launch. "Having someone from the field you're trying to solve come ask you for your product and get excited by it," Taavi says, "is a different level of motivation."
The Hard Part
The near-death moment was the engine failing on legal vocabulary. Thousands of pages in, and it simply didn't understand the terms that had no plain-language definition. The output was off. A month's work looked like it had gone down the drain.
What followed was less a fix than an education, both for the agent and the team. They picked themselves up and started teaching the model law, properly, term by term. Many test-runs later, the verdict is that it was worth it: the failure forced them to understand their own engine far more thoroughly than success would have.
It also corrected the assumption they'd started with. They'd believed LLMs were already advanced enough to reason about legal questions out of the box. The more they trained, the more they understood how wrong that was and how complex the mission actually is. The lesson Taavi draws is sharper than "AI is hard": educate yourself in every inch of one field and the methods involved and bring in people from that field who have real-world knowledge, because they'll pinpoint the exact problems and opportunities that outsiders can't see.
What's Next
In three years, the goal is for Juvo Legal to be embedded across sectors and law firms as a core function of how legal work actually gets done, not just a tool bolted on the side.
They're clear-eyed about geography. Expansion into neighbouring countries is a real possibility, but Taavi doesn't kid himself about the sequence: get a fully working system to market first, then think about crossing borders. "The road ahead is long," he says, "but we are excited to take on the challenge."
The Toolbox
Juvo Legal runs on what Taavi calls the three C's: Claude, coffee, and calmness. The first two are self-explanatory. The third, he insists, isn't a joke as staying calm through the difficult moments and holding a clear vision of what's ahead is genuinely vital when you're building.
For anyone who needs it, his recommendation isn't a book. It's Steve Jobs' 2005 Stanford commencement speech, the perfect articulation, he says, of the fact that the journey will have its ups and downs, and every opportunity opens a door somewhere new. Keep exploring.
⚖️ How you can help: Juvo Legal is looking for its first real feedback from people living the problem. If your legal-compliance process is taking too long, or you're not confident your terms of service are airtight, Taavi wants to hear from you, the conversation is likely to help both sides. Reach out via the Linkedin pages of Taavi, Georg and Kristen.
Written by
Mihkel Hõbemägi · Co-founder, Built in Baltics