Dear readers,

This Wednesday we take a look at how AI could reshape one of the world’s most information-heavy industries. We sat down with Infobip co-founder Roberto Kutić, who is now building Lexandro AI with a bigger ambition than simply putting a chatbot in front of lawyers: turning fragmented legal knowledge into something businesses and eventually everyday users can actually use.

Across the SEE ecosystem, we also look at Forty.5 Ventures’ new $30M fund targeting early-stage startups in the region, Romania’s Veridion raises a $20M Series A round, and Transylvania Angels Network’s €50K investment in The Founder Games.

Additionally, Rumor Has It that while SEE founders say VC capital is out of reach, investors argue the bigger problem is a lack of genuinely venture-scale ideas. And in our Ecosystem Take, Vestbee’s Lisa Palchynska explains why regional startups can be technically strong but still struggle to turn that excellence into stories that connect with investors, customers and the media.

Enjoy the read!
Bojan Stojkovski
Editor-in-Chief, IT Logs

If you take a look at his career, you could see certain inevitability to Roberto Kutić’s move from communications technology into legal AI. The Croatian veteran entrepreneur has spent decades watching technology move through several distinct phases. First came personal computers, then the internet and the dot-com era. Enterprise software followed, connecting companies and their data through systems such as CRM and ERP, and then came platforms.

Now, Kutić says that we are in the AI phase of it all. But he is less interested in AI as a futuristic technology than in what happens when it is applied to industries that have never been properly digitized. That is the thinking behind Lexandro AI, a legal-tech company Kutić is building after helping scale Infobip, the Croatian communications technology company that became one of the region’s most important tech success stories.

Kutić describes his new focus as an attempt to tackle something he and his colleagues experienced themselves: the frustrating, fragmented and highly manual interaction between people, companies and the legal system.

The ambition, however, goes beyond building another chatbot for lawyers, as Lexandro is being developed around a more fundamental problem: how do you take enormous quantities of legal, corporate and institutional knowledge, structure it properly and make it usable without compromising security, privacy or context.

The next AI race is about context

Kutić is hardly unaware of the competition, pointing out companies such as Legora and Harvey as examples of startups that have built strong products and brands around using large language models to work with legal knowledge.

On the other side of the Atlantic, he was watching Harvey. The main idea across much of the sector is similar: take the knowledge that exists inside a law firm or enterprise and make it accessible through AI.

However, Kutić argues that the competitive advantage will not necessarily come from whoever has the biggest or most fashionable model. It will come from what happens before the model generates an answer. “The most important part is the preparation of the context,” he said. “LLM, yes, you need to have a good LLM, this is important, but not so important like the preparation of the data.”

That is a deceptively difficult engineering problem. A lawyer's case may contain hundreds or thousands of pages. Contracts, previous decisions, correspondence, regulations, tax information, scanned documents and other records may all be relevant. A conventional AI system can and most likely will struggle when too much of that information needs to be considered simultaneously.

The company's approach also includes chronology, allowing users to move backwards through a case and understand what information was available at a particular point in time. For lawyers, that can be more useful than simply asking an AI model to summarize a document.

Legal AI cannot simply be imported

There is another problem with applying AI to law: law is local. A model trained primarily on US or UK legal material cannot simply be dropped into Croatia, Serbia or another Balkan market and expected to understand the nuances of the local legal system.

That is why Kutić sees localization as a feature rather than a limitation. “Law is everything about confidentiality, but also about privacy. And altogether, you need to be local to be good,” he says.

The idea is not merely to put local case law into a database. Kutić sees a broader opportunity to connect different types of information that businesses and lawyers need, including tax advice and other sources relevant to a particular jurisdiction. The company’s strategy is deliberately market-by-market, with local partners playing an important role, partly because of the nature of legal information itself.

For all the excitement around AI in legal services, lawyers have an unusually obvious reason to be cautious. Their systems contain confidential client information, privileged communications and sensitive corporate documents. Sending that information into a generic public AI service is therefore a fundamentally different proposition from asking a chatbot to write a dinner invitation.

Kutić says security was one of the three major issues lawyers raised when he began speaking to potential customers. “The first and foremost is security,” he emphasizes.

Lexandro was built with Microsoft's Azure infrastructure from the beginning, and Kutić says the company is taking GDPR and the EU AI Act into consideration as part of the product architecture.

He adds that Lexandro is also designed to ground its answers in source material. “Our system is grounded, so we are never giving out any information that you can’t prove,” Kutić said. “So everything comes with sources, and everything has a citation.”

The second major issue is the sheer amount of knowledge the system has to process, while the third is speed. Kutić argues that lawyers need the system to be able to process large quantities of information while still delivering a usable, real-time interaction. Lexandro is being developed with customers directly involved in that process.

“We have a few customers that are helping us to shape the product,” he tells IT Logs. That has already resulted in additional functionality, including a Word plugin, while the company is continuing to develop integrations with enterprise systems such as SharePoint and Google.

The lawyers may be more ready than expected

While Kutić admits he was initially skeptical about how quickly lawyers would embrace AI, he has since changed his mind. “I was totally skeptical, but they are very, very keen for new things,” he said.

That does not mean lawyers want AI to replace them. In Kutić's view, the opposite may happen: AI could shift the profession toward the parts of legal work where humans remain most valuable.

He sees legal work as having several layers. Lawyers need to absorb enormous amounts of information and apply logic to build a case. Then comes strategy, negotiation and human interaction. AI can increasingly help with the first part, while the second may become even more important.

That includes understanding a client, negotiating with the other side and finding common ground. It is also an area where Lexandro is experimenting with AI tools that can help lawyers understand the other party in a dispute and anticipate what they might want or do next.

The irony is that better AI could make human skills more valuable rather than less. Furthermore, Lexandro itself  is part of a broader structure. Kutić says his new startup Festina is focused on AI knowledge and can work with companies outside the strictly legal environment. 

The same underlying technology can be applied to other information-heavy industries. One example is a regional pharmaceutical company, where Kutić says a pilot is being prepared to help with compliance involving pharmaceutical products and regulations in both the Western Balkans and the EU.

Tasks that previously required employees to spend days looking for similar pharmaceutical components and carrying out comparisons could potentially be reduced dramatically. That is where Kutić sees a much bigger market emerging. “If some company needs AI knowledge, then we go there, we analyze what they need and we help again with the Lexandro product,” he said, adding that his longer-term vision is essentially an AI layer sitting on top of enterprise knowledge.

The evolution, as he describes it, is straightforward: first companies digitized their operations, then connected their departments and data. AI now provides another layer on top of that infrastructure, allowing organizations to interact with their accumulated knowledge.

And eventually, he expects AI agents to start acting on that knowledge. Lexandro already has dozens of agents and engines inside the platform, according to Kutić, with further automation of business processes planned.

Democratizing the legal system

Perhaps the most ambitious part of the project is not aimed at lawyers at all. Kutić wants Lexandro eventually to offer smaller, more accessible packages for ordinary users. Legal services remain expensive and difficult to access for many people, particularly when the issue does not justify hiring a lawyer at traditional rates. “I think this democratization will come, but by digitalization,” he said.

The first major shift, he expects, will happen inside companies, where in-house legal teams can use AI to connect information from different departments and deal with routine legal work more autonomously.

But eventually, the same technology could move further down the market. Simple legal questions, contracts, documents and other everyday problems could become increasingly self-service.

That does not mean the lawyer disappears. Instead, Kutić expects AI to handle more of the information-heavy work while humans focus on judgment, strategy and negotiation. According to the Croatian entrepreneur, law is full of information - but it is also full of outdated processes, fragmented archives and local rules.

That may be an unusually simple philosophy for a complicated AI market. But after Infobip, Kutić is betting that the same discipline that helped build a major tech company can be applied to a very different kind of problem: turning the world's messy legal knowledge into something machines, lawyers and eventually everyone else can actually use.

IT Logs:  What separates a compelling startup story from a good marketing pitch?

IT Logs: Do startups from SEE struggle to tell their stories differently than founders in Western Europe or the US?

IT Logs: Why is translating technical excellence into a compelling narrative still a challenge for SEE startups?