Becoming a Research Institution in Applied AI

The Illusion of Certainty

Artificial intelligence systems increasingly influence decisions that carry real consequences — from financial assessments to policy analysis and institutional communication. In many environments, algorithmic outputs are no longer treated as exploratory signals; they are read as answers.

The speed and seamlessness of these systems create an impression of clarity. Dashboards update instantly. Summaries appear coherent. Classifications arrive with confidence. Yet beneath that clarity lie assumptions about data selection, model design, and evaluation criteria. When those assumptions remain invisible, uncertainty itself becomes invisible.

The problem is not that AI systems are imperfect. All analytical systems are. The deeper risk is misplaced certainty — the tendency to treat probabilistic outputs as definitive conclusions.

In applied contexts, where decisions shape institutions and markets, the distinction between experimentation and infrastructure becomes critical. Building quickly is not the same as building responsibly. Generating an answer is not the same as validating it.

Research discipline exists precisely in that gap.

For Event Registry, becoming a research institution formalizes an approach that has long guided its work: applied AI must be grounded in methodological rigor and a clear separation between exploration and production.

To understand what this means in practice, we spoke with Erik Novak, PhD, who leads research and data science at Event Registry, and with CEO Gregor Leban, PhD. Both are formally registered researchers within Slovenia’s national research framework.

Research in Practice

When asked what it means, in practical terms, for a company like Event Registry to operate as a research organization, Erik Novak begins by looking beyond the formal designation.

Beyond the formal title, he explains, it means that Event Registry is dedicated not only to using existing tools, methods, and algorithms developed by others, but also to being a creator of such tools, methods, and algorithms itself — building on the work of other researchers and sharing its own work with the research community.

It also serves as a promise: that the company will try to build better solutions and continuously improve its algorithms, services, and systems from a scientific standpoint — not because this is new, but because it formalizes an approach that has already been part of its work.

Misunderstanding AI Systems

Today, Novak most often sees AI systems misunderstood in the context of large language models and AI agents.

The rise of LLMs has brought significant opportunities in both academia and industry. But it has also introduced traps into which people too often fall — such as assuming that what an AI system returns is absolute truth, or that AI agents are always fetching the correct data or information.

In reality, these systems operate probabilistically and depend heavily on the quality and structure of the data they access.

This is where data-driven systems play a stabilizing role. Systems such as NewsAPI already curate data, enrich it, and expose it in structured form through an API. AI agents can then operate on this curated layer — provided they are properly integrated.

In this sense, AI agents and data-driven systems can work in tandem. One supports the other. The reliability of downstream systems depends on the quality, structure, and governance of the upstream data layer.

Transparency and Its Limits

For Novak, transparency is fundamental.

If there is confusion about how a system works or how it returns information, decision-makers can misinterpret outputs and make consequential mistakes. That is why it is critical to make systems, algorithms, and methods as transparent as possible — including through peer-reviewed scientific publications, which provide credibility to the underlying approaches.

At the same time, industry operates within a competitive landscape. Researchers must decide which results can be published openly and which are better kept proprietary. This includes difficult choices about whether to release datasets or models, in an environment where today’s research community often expects openness and open-source contributions.

Navigating between academia and industry is therefore a challenge in itself. It requires judgment — not only about what is scientifically sound, but also about what is strategically sustainable.

Experimentation and Responsible Integration

From a research perspective, the difference between building AI systems quickly and building them responsibly is substantial.

Building quickly allows teams to test ideas, experiment with methods, and create new features rapidly. But speed can make developers unaware of pitfalls. There may not be enough time to cover all possible edge cases, to rigorously test and experiment with new solutions, to consider how those solutions should be optimally integrated into existing systems, or even to validate whether the feature addresses a real demand.

Novak describes this as a separation between two phases.

The first is research: quick and sometimes messy experimentation and prototype creation. The second is innovation: a slower, more thoughtful process that examines all angles of integration into the AI system and evaluates whether the solution truly addresses a problem worth solving.

Maintaining this separation is essential. Without it, experimentation risks becoming part of the core system before it has been sufficiently tested, validated, and understood.

Event Registry is officially registered as a research organization with ARIS — the Slovenian Research and Innovation Agency.

Responsibility in Interconnected Systems

While research discipline defines how systems are built, responsibility defines how they are positioned in the world.

For Gregor Leban, the question is not only whether an AI system performs well in isolation. It is how it behaves when connected to other systems — and how its outputs influence real decisions.

“When AI systems interact with other systems,” he explains, “uncertainty compounds. Assumptions from one layer can quietly propagate into another. If the data structure is ambiguous, if governance is unclear, or if interfaces are loosely defined, small inconsistencies can scale.”

This is where interoperability and structured data governance become critical.

Interoperability is not merely a technical feature. It is a discipline of clarity. When systems are designed to communicate in structured, well-defined formats, ambiguity is reduced. Downstream systems can trace how information was produced, filtered, or transformed. The chain of reasoning becomes inspectable.

Structured data governance plays a similar role. It ensures that information is not only available, but organized, contextualized, and documented in ways that allow interpretation without guesswork.

“In high-stakes environments,” Leban continues, “you cannot rely on systems that behave like black boxes. Decision-makers need to understand not only the result, but the process behind it. That requires intentional design.

As AI systems become infrastructure, their influence expands. When influence expands, discipline becomes more important, not less. Speed is valuable. But reliability and clarity determine whether a system can be trusted over time.

When decisions depend on these systems, responsibility is no longer abstract.”

The distinction is subtle but consequential. Research discipline governs how solutions are created. Responsibility governs how they are embedded.

Together, they define an approach that prioritizes long-term stability over short-term acceleration.

Institutional Recognition

Event Registry is officially registered as a research organization with Javna agencija za znanstveno raziskovalno in inovacijsko dejavnost Republike Slovenije (ARIS) — the Slovenian Research and Innovation Agency.

ARIS operates within Slovenia’s national research and innovation framework, overseeing the evaluation and classification of research organizations and researchers. Through the national SICRIS system, research activities, fields, and outputs are formally documented and publicly traceable. As part of its mission, ARIS promotes internationally comparable evaluation standards and supports research cooperation across borders, aligning Slovenian research activity with broader European frameworks.

The registration places Event Registry within this formal research structure, with recognized research areas including algorithms, optimization, and intelligent systems.

The designation does not redefine the company’s direction. Rather, it formalizes an orientation that has guided its work: approaching applied AI not only as engineering, but as research practice.

Discipline as Ongoing Commitment

Becoming a research institution is not a transformation achieved in a single administrative step. It is a commitment to a particular way of working.

It means accepting that experimentation and production are not the same. It means acknowledging that uncertainty must be visible, not concealed behind polished outputs. It means understanding that as systems scale, their consequences expand.

For Novak, the most compelling aspect of working at the intersection of applied research and real-world systems lies precisely in this tension.

“What excites me the most,” he reflects, “is that what we are building is helpful for so many people, both in academia and industry. Sometimes the solutions are straightforward. But many times, they require research — finding whether a solution already exists and, if not, building it from scratch.”

That process rarely follows a linear path. It requires adapting methods to the speed and volume of data being processed. It requires recognizing that large language models are not always the right tool for every problem. And it requires building systems that can, in turn, become the right infrastructure for those models — structured, governed, and capable of supporting them reliably.

The distinction between speed and discipline is unlikely to disappear. If anything, it will become more pronounced as AI systems continue to integrate into institutional workflows and decision-making processes.

In that environment, research is not a status. It is not a signal of prestige. It is a practice — one that demands patience, scrutiny, and humility.

Applied AI will continue to evolve. Methods will improve. Tools will change. But the principle remains constant: when systems influence real decisions, discipline must precede acceleration.

Becoming a research institution formalizes that principle. Sustaining it will remain the ongoing work.