Egzakta co-founder and partner Nenad Tešić is one of the three founders behind the company, alongside Zoran Radisavljević and Marko Marković. In an interview for BIZLife, he highlights that the company is backed by a team of 200 experts across four countries, a client portfolio that includes the largest banks, telecom operators, and government institutions in the region, as well as an AI product that is gradually becoming a benchmark for sovereign artificial intelligence in regulated industries.

Egzakta has been around for more than 10 years. How did you get to a world-class AI product?

Through evolution, not revolution. The development followed its own internal logic, which today seems almost inevitable - although at the moment individual decisions were made, it did not always appear that way.

We started as management and financial consultants. We helped companies understand where they were, where they wanted to go, and how to get there - strategy, organization, financial planning, restructuring. But over time, we noticed a pattern: once a client understood where they wanted to go, the next question was always how to execute it in practice. And more often than not, the answer was that IT had to change as well. Processes, systems, architecture - everything. That was the natural next step. If you show someone what something should look like, it is logical to help them build it too. That is how management consulting organically evolved into IT consulting and implementation. One discipline kept opening the door to the next. It also became increasingly clear that IT was no longer just a cost center, but the very heart of every company.

At the same time, as a Group, we were also building an investment pillar - acquiring companies in related industries, learning how they functioned internally, improving operations, and preparing them for further development or exit. It was a way for our team to gain a much deeper understanding of business models that consultants usually observe only from the outside. Every acquisition taught us something. Some things went well, others less so, but each one provided knowledge you cannot get from books or consulting projects alone.

When you combine nearly two decades of consulting across almost every industry, IT implementations in the region’s most critical institutions, and operational experience gained through directly managing and developing companies - you get a team that understands business in a way pure technology companies rarely can.

That understanding ultimately led us to AI. We did not enter artificial intelligence as technology enthusiasts looking to experiment with something new. We entered it as people who had spent years observing the same problems across clients and realized that AI finally offered tools capable of solving them in ways that had not previously been possible. The question was never whether we should work with AI - that was obvious. The real question was whether we would become another reseller of other people’s solutions or build something of our own. We chose the latter, and that path led us to KVARK.

When exactly did your AI journey begin, and what did it look like?

We were not Turing, nor were we the first AI laboratory at some prestigious university. It all started quite modestly around three years ago, when we began experimenting with tools to create marketing materials for our brand Revita. Image generation, marketing visuals - nothing spectacular, but something clicked at that moment.

That aligned with Marko’s vision - he recognized that this was something far deeper than image generation and launched the AI Lab as a dedicated research branch. We started systematically exploring open-source tools and integrating them into everyday work. The approach was pragmatic: how can this accelerate what we are already doing?

In practice, consultants began using language models to prepare documents, analyze materials, and conduct market research - tasks that once took days were reduced to hours. Project managers used AI for reporting automation and status tracking. Analysts significantly accelerated and improved the reliability of data work while reducing the margin for error. Gradually, AI stopped being a separate project - it became part of the way we work, embedded into every team.

Then the entire development division began widely adopting AI tools in day-to-day operations. AI-assisted code completion, code review, automated testing, documentation generation. The result was concrete and measurable - development team productivity increased by more than tenfold. Not as a marketing slogan, but as a real change in how much code the team could deliver in a single iteration.

Then came the next step - the critical one. We realized that no one in the market had moved in one specific direction: building an AI orchestration platform designed for enterprise-scale systems, with all their complexity, regulation, security requirements, fragmented data, and sophisticated IT architectures. Everyone was selling models or cloud access. Nobody was solving deployment inside institutions. That realization became the foundation of KVARK.ai.

People often talk about AI agents as the next major wave. What are they really, and why is orchestration so important?

This is something Marko explains much better than I do, and I have to admit one of his analogies completely changed the way I see it:

“Imagine a restaurant with twelve chefs. Each one extraordinary in their specialty - one makes the perfect risotto, another sushi, a third bakes bread that smells like childhood memories… But there is no head chef. No coordination. No menu concept. The result is not dinner. The result is expensive chaos.”

AI agents are becoming operational actors in business. They perform increasingly complex tasks autonomously and without human supervision. They begin making decisions independently - even decisions they were never authorized to make. They start creating their own agents, building their own network of workers operating 24/7. I know this now sounds a bit like the origin story of Skynet from The Terminator - one of the most famous cultural symbols of fear surrounding artificial intelligence. Without a control layer that governs them, defines the limits of their autonomy, and records every decision, it becomes dangerous.

That is where KVARK.ai comes in - as the control system that keeps all those agents coordinated and within the boundaries defined by the company.

Statistics show that 80–95% of AI pilots fail to generate ROI. Why are so many AI projects unsuccessful?

Because companies are still buying technology instead of solutions - unfortunately, in most cases. They attend a presentation, get impressed by the demo, sign a contract for a cloud AI tool, and only later realize they cannot actually provide it with their data because regulation prevents it. Or they start calculating costs under a per-token pricing model and realize the monthly bill no longer matches the original business case. Then the CFO enters the conversation demanding cost predictability and refusing uncontrolled spending. Or they discover the AI cannot access data trapped across five disconnected systems.

We identified six main causes of AI project failure: data privacy, unpredictable costs, siloed data, uncontrolled employee use of public AI tools, regulatory risks, and unrealistic expectations. KVARK was designed from day one to eliminate all six, while our implementation methodology ensures that transformation actually reaches production instead of remaining stuck at the pilot phase.

What exactly is KVARK, and how is it different from what already exists on the market?

There is a clear trend today - every company and institution feels pressure to adopt AI. The problem is that everyone runs into the same barriers. Data is confidential, GDPR exists, the EU AI Act is coming, and the NIS2 Directive increases security pressure. Yet most major AI vendors still offer cloud-only solutions. The gap between what the market offers and what regulated institutions can realistically adopt is enormous.

KVARK fills that gap. KVARK is an Enterprise AI Factory that operates entirely within the client’s own infrastructure. To simplify it: we bring in the server, connect it to the client’s network, integrate it with everything they already use - ERP, CRM, SharePoint, databases, archives - and make all of it accessible through a simple interface. Data never leaves the building. The client owns the model, the data, and the infrastructure. They are dependent on nobody. Companies that build their own orchestration layer - model-independent, locally hosted, and fully auditable - retain control over their AI future.

Starting August 2, 2026, the European Union will begin penalizing companies that use AI without compliance with the new law. Fines can reach up to 7% of global annual revenue or €35 million. What many still do not understand is that the responsibility falls not on software providers, but on companies using AI in high-risk areas such as hiring, lending, or healthcare. The requirements are concrete: human oversight, risk assessment, transparency, and audit trails. Most SMEs assume their vendor handles compliance - the Act says otherwise. As with GDPR, companies that prepare early gain an advantage; those that wait end up paying the price.

What does implementation actually look like - from the first meeting to deployment?

This is perhaps what differentiates us most from a market that tends to sell AI as a plug-and-play product. Our approach is methodological and structured into four distinct phases.

We begin with a pre-project phase - understanding the context, involving key stakeholders, and defining the pace of work. Then comes business analysis, which is the most important phase for us: we map existing processes, identify gaps, and only then design the target state. This step follows the principles of a true consulting engagement and provides clear insight into needs and possible solutions. Without it, writing a single line of code makes no sense.

Implementation takes place in two-week cycles - each delivers concrete, tested functionalities that the client approves before we move forward. Nothing is sent to user acceptance testing until it is logically complete and fully functional. Finally comes support - not as a formality, but as a feedback loop through which we fine-tune the user experience in production.

The result is that the client always knows exactly where we are and what they are getting. There are no surprises at the end of the project.

Can you share concrete examples - what are your biggest projects and what results have they delivered?

We work with institutions whose results are difficult to ignore, in industries where mistakes are not an option - public healthcare systems, pension funds, ministries, banks, telecom operators, FMCG companies, and more.

Behind us are more than 300 projects with measurable outcomes: systems that reduced processing times from weeks to hours, platforms that increased revenues by double-digit percentages, and digitalization initiatives that transformed paper-based processes into automated workflows operating 24/7. We have worked with leading regional banks on core banking transformation projects, with government institutions on systems directly affecting the daily lives of millions of citizens, and with multinational companies on process optimization across multiple countries simultaneously.

Every project shares one characteristic: long-term relationships and measurable results. It is no coincidence that 70% of our clients return or recommend us to others.

Besides KVARK, you also have TubeIQ, an early warning system, electronic document management, an electronic parliament… How do you manage such a broad product portfolio?

Through synergy. What others might see as a complicated portfolio, we see as an ecosystem where every component has a role and supports the others.

TubeIQ is the digital nervous system - process orchestration, workflow automation, document and digital archive management, digital onboarding for banks and telecom operators compliant with KYC/AML regulation, digital signatures, procurement management, HR modules, early warning systems for financial institutions, public debt management, market inspection, legal collections - and the list goes on. Everything originated from the same principle: the need for digitalization. Before AI can be introduced into processes, those processes first need to be digitalized. AI cannot learn from chaos. TubeIQ handles that first step, while KVARK places AI on top of it.

All of these products emerged from concrete client demands. We did not invent them in a vacuum - we identified problems in the field, developed solutions, and then offered them to the next client facing the same issue.

You also have LM TEK behind the EK liquid cooling brand. How does hardware fit into your story?

That is a key differentiator separating us from most AI software companies. LM TEK is a Slovenian company whose operational control we assumed in March 2025. It stands behind the global EK brand - a world leader in liquid cooling with more than 15 years of history and a worldwide distribution network.

EK has two clearly separated directions. The first is EK Quantum Cooling - the consumer segment focused on premium liquid cooling components for personal computers, long regarded as a reference brand among enthusiasts and professional gamers worldwide. The second, strategically far more important for us, is EK Fluid Works - a division focused on liquid cooling for AI servers and high-performance computing centers using direct-to-chip technology. This enables GPU density levels impossible with traditional air cooling.

The combination of EK Fluid Works technology and KVARK software resulted in the RM-4U8G server - a rack server with up to eight NVIDIA H200 GPUs, liquid-cooled, forming the hardware backbone of the KVARK Enterprise AI Factory.

And this is where our uniqueness becomes clear. When one company combines its own liquid-cooled hardware platform, its own AI orchestration software, and twenty years of consulting experience - you get something that barely exists on the global market. Around the world, there are companies that do parts of this. We do everything, from the first client conversation to the server operating inside the client’s data center.

What does the market math show - why is owning infrastructure more cost-effective than the cloud in the long run?

This is a topic people rarely discuss openly, yet it is increasingly becoming the decisive factor when clients choose between cloud-based AI and on-premise infrastructure.

Cloud AI comes with several hidden costs nobody mentions during sales presentations. The first is usage-based billing - every query costs money. During the pilot phase, this seems insignificant. In production, with hundreds of daily users, monthly bills become unpredictable and often shocking. Companies that entered cloud AI with pilot budgets suddenly faced costs multiple times higher once systems were widely adopted.

The second hidden cost is dependency. When you build business processes around someone else’s cloud model, you become a tenant. And the landlord can change the rules whenever they want - pricing, availability, functionality. This is not theoretical; it is already happening. Companies that built on specific cloud AI services discovered that conditions changed, and migrating elsewhere became more expensive than expected.

The third cost is harder to measure - dependency on the vendor’s engineering team. Every customization, integration, or modification requires vendor involvement. Your IT department becomes a coordinator instead of an owner.

We conducted a concrete five-year analysis for organizations with more than 100 users. KVARK running on private infrastructure costs 86% less than an equivalent cloud solution over that period. These are not estimates - they are real numbers from real projects. And prices are expected to continue increasing.

For example, once a company reaches 50 million API calls per month, the cost gap between cloud and locally hosted infrastructure becomes dramatic.

Cloud solutions in the GPT-5.4 class cost between €2.5 million and €4 million annually - and costs continue to rise with every new use case. By contrast, open-weight models such as DeepSeek V3.2 or Qwen3 235B achieve similar performance while running on private infrastructure for between $400,000 and $700,000 annually. With a one-time investment of around $300,000 for an orchestration layer, ROI is reached by month 14, and from year three onward, savings exceed €2 million annually - permanently.

The CFO knows exactly what is being paid and when. All models and everything the system learns remain the client’s property. No vendor lock-in, no pricing surprises, no unexpected invoices at the end of the month.

Your team consists of 200 experts, average age 34, and 53% women. How do you retain talent in Serbia?

This may be the hardest and most important question in the entire conversation. Everything we have built - KVARK.ai, TubeIQ, our projects, our clients - none of it exists without the people behind it.

Generation Z now makes up a significant part of our workforce, and it would be dishonest not to admit that they have completely different expectations from what my generation considered a “normal job.” Work-life balance is not a perk for them - it is a basic condition. Meaningful work matters just as much as salary. At first, I admit, that was not always easy for me to understand. I grew up in an office culture built around presence, personal relationships, and physical attendance. But I learned that if you do not adapt, you lose your best people.

What we realized is that management based on command and control does not work for the kind of people we attract and want to keep. A young engineer capable of solving an architectural problem in hours that might take a senior engineer days does not want a manager supervising every step. They want challenging problems and the freedom to solve them in their own way. That is why we built a culture where responsibility comes quickly. If you are ready, you do not wait. In fact, 36% of our employees started their careers at Egzakta - not by coincidence, but because young people see that growth here is not a waiting game.

When AI entered everyday operations, something interesting happened. Younger team members did not wait for instructions - they were the first to experiment, learn, and apply new tools. That led us to adopt what the world calls reverse mentoring - younger colleagues educating older ones about new tools and workflows. For us, this is not a formal program; it happens naturally every day. A senior consultant learns from a younger engineer how to use AI in preparing an analysis. That may be one of the best things that has happened inside the company in the last few years.

This combination of experienced professionals who brought knowledge from outside and people who grew within the company is what gives our organization strength. It is difficult to replicate. Our workforce is 53% women, 92% of employees hold university degrees, 35% have MBAs, and the average age is 34. It is a team capable of speaking with both a finance minister and a systems architect in the same morning - and understanding both equally well. That is our real advantage.

Where do you see Egzakta in five years?

I see a company that has become a reference point for sovereign AI deployment across Europe and the Middle East. Not as a vendor selling tools, but as a strategic partner delivering full transformation - from consulting and implementation to AI factories that remain fully owned by the client.

Honestly, as much as I care about where we will be in five years, I care equally about how we get there. The pace of technological change is such that the skills you have today will not be enough five years from now - perhaps not even sooner. That applies to our clients, but also to us. A company that stops learning stops being relevant. That cannot happen to us, and I do not believe it can happen to any serious company.

Finally, how do Harley-Davidson and sovereign AI infrastructure coexist in one person?

It is difficult to explain to someone who has never been infected by that virus. And virus really is the right word - once it catches you, there is no cure.

Something happens when you sit on a Harley, start the engine, and hit the road. The sound, the vibration, the road ahead. In that moment, emails disappear, pending decisions disappear, delayed projects disappear. It is a kind of freedom very few things in life can provide. It is the adrenaline every rider seeks, even if they would not always describe it that way.

But there is another side to it that may surprise people unfamiliar with the Harley-Davidson culture. Harley is not a motorcycle for people who want to take risks - it is for people who want to arrive. Massive, stable, built to last. When you have three boys at home waiting for their father, that stability matters. Every time I ride, I know this is a machine that punishes recklessness but protects you if you respect it. There is a philosophy in that - freedom and responsibility are not opposites; they can coexist.

My three boys look at that motorcycle with a mixture of fear and fascination. In the same way, I look at some of the business decisions we have made. You know there is risk involved, you know it demands respect, but you also know why you are doing it.

People infected by the same virus will understand this without further explanation. Those who are not - will not, and that is perfectly fine. Some things are not for everyone, and that is exactly what makes them valuable.

There is also a business parallel, and it is not forced. Harley is a symbol of independence. We sell digital independence - to companies, institutions, and governments that do not want to depend on someone else’s AI platforms and servers. Those who fail to build their own AI infrastructure become dependent on those who do. That is the geopolitical reality of the 21st century, just as energy dependence defined the 20th. How many companies, institutions, and governments realize that in time remains to be seen.

And there is one more similarity. Harley is not a motorcycle you ride without experience. It is not your first bike, nor your second - it is a motorcycle you choose after you have ridden everything else. After you have fallen, learned, accumulated miles, and understood what it means to respect the machine beneath you. Experienced riders know this. Young riders who jump onto a Harley too early quickly learn that the bike does not forgive mistakes.

AI is exactly the same. Today, everyone wants to jump directly onto the biggest machine without knowledge, experience, or understanding of what happens under the hood. Then they wonder why pilots fail, why costs explode, or why systems make decisions nobody predicted. Neither AI nor Harley forgives ignorance.

What we bring to clients is not just technology - it is twenty years of experience in understanding how institutions actually function, what slows them down, and what they truly need. You cannot buy that during a sales presentation. It is earned only through work, project by project, year by year. Just like miles on a motorcycle - there are no shortcuts.