AI2Future 2026

Check out the Schedule

AGENDA

Day 0
Day 1
Day 2

Stay tuned for 2026 updates!

Main Stage
Business & Implementation Track
Engineering Track

09:30-09:32

Conference Opening

09:32-09:35

Valentina Zadrija

AI2Future 2026 Programme Committee; CroAI Board Member

Valentina Zadrija

Director of AI at Gideon

Valentina is the Director of AI at Gideon, where she develops AI models for robotics and physical AI. She has over 20 years of experience in the field, working at the intersection of LLMs, computer vision, and 3D understanding models. Valentina obtained her PhD in AI from the University of Zagreb in 2017, focusing on weakly supervised learning and traffic safety. After completing her PhD, she joined Rimac Automobili to work on scene-understanding models, and later moved to Gideon to contribute to autonomous mobile robotics. She spent the last few years at Yaak AI GmbH, where she trained VLA models and contributed to the HuggingFace LeRobot platform. She has now returned to Gideon, working to make robotics in Croatia great (again). Valentina is a mother of three and serves as an ambassador for the Female Engineer of the Year initiative in Croatia.

09:35-10:10

Arsène Lavaux Santonacci (SilenceSilence.ai)

Human art, robotic science, and the agents that only work when both are in the system

Arsène Lavaux Santonacci

Founder & CEO at SilenceSilence.ai

Arsène Lavaux Santonacci is the founder and CEO of SilenceSilence.ai, where he is building a Clinical Human Voice Restoration OS to help people who lose their voice preserve and restore what makes it uniquely their own.

The project began with a promise to his childhood friend Richard, who lost his voice to ALS. That personal commitment grew into a broader mission to make personalized voice restoration accessible to people affected by conditions that impair speech. Clinical studies are underway in Europe, with plans to expand to the United States.

Since 2023, Arsène has also deployed agentic AI solutions in healthcare and enterprise environments.In the United States, he worked with the Purchaser Business Group on Health (PBGH), developing next-generation healthcare services for members representing major global employers.

Alongside his work in health AI, Arsène has advised founders across the US, Europe, Asia and Africa, including entrepreneurs from Y Combinator, Techstars and 500 Global. He is a growth mentor for 500 Global and HEC startups and co-founded the French chapter of the Agentics Foundation.

10:10-10:40

Abhishek Bhargava (Datum)

From Demo to Deployed: What Actually Gets AI Into Production

Abhishek Bhargava

Co-founder at Datum

Abhishek Bhargava is the co-founder of Datum, an AI software and services company building a suite of AI products and delivering custom AI solutions for enterprise clients across industries. At Datum, he leads the work of translating frontier AI capabilities into production systems — spanning agentic workflows, voice agents, and domain-specific applications — and partners directly with clients to architect, deploy, and scale these solutions in real operational environments.

Before Datum, Abhishek founded Vectordash (Y Combinator W19), where he built low-latency GPU streaming technology that put real-time computer vision workloads in the cloud — a thread that runs through much of his career. He's shipped production ML and computer vision systems as an engineer at a robotics startup, conducted deep learning research at MITRE, and built quantitative trading infrastructure at a commodities hedge fund. He later launched his own crypto hedge fund, deploying strategies that outperformed the market. His published research spans bioinformatics (using data science to design HIV-inhibiting peptides), NLP (duplicate question detection with word2vec and CNNs), and reinforcement learning applied to equity markets using probabilistic guarantees from cyber-physical systems — a strategy that significantly outperformed the S&P 500 over a 15-year backtest.

Abhishek studied Computer Science, Computational Finance, and Robotics at Carnegie Mellon. At Datum, he focuses on turning deep technical expertise into AI implementations and partnerships that move the needle for clients.

10:40-11:10

Irja Straus (Jungheinrich)

Your Evals Are Tests. Now What? AI made everyone a tester. The craft already exists.

"Looks good to me."

"It passed the eval."

"Just run it again."

"I don't care as long as it's green."

Let's be honest, if you work with AI, you've said or heard at least one of these this month. I heard the last one from product owners when I asked what they thought about our test automation. Surprise, surprise, green means nothing if nobody checks what it measures. Every time you decide if something is good enough to ship, you are testing, whatever your job title says. An eval is just a test.

I've tested AI from both sides. I led testing for face recognition models, where the AI was the thing under test. Today I test yellow electric forklifts, where AI tests with us: predictive models replace part of our physical battery cell tests. Which brings me to my question: when a model replaces a test, who tests the model?

In this talk, I will share how hardware-in-the-loop test rigs and cell testing work together with predictive models, and how we decide when a model is good enough to skip a physical test. Then we'll look at some techniques you can use on any AI system, such as metamorphic relations (what must stay the same when the input changes). You will leave with a better answer to "is it good enough?" than "it's green".

Irja Straus

Team Lead Testing Electronic Components at Jungheinrich

Irja Straus leads testing for embedded software and electronic components at Jungheinrich Croatia, and is the founder of a consulting company specializing in quality engineering and testing. She began her career as a software tester and business analyst, later moving into a product manager role. After several years in product, she took on the challenge of establishing testing practices and embedding quality from the earliest stages of development. For the past few years, she has focused on contemporary exploratory testing, building quality engineering strategy and effective cross-team collaboration.

11:10-11:40

COFFEE BREAK

11:40-12:10

Desislava Sarbinovska (Service Design Digital) & Luka Baranović (Humanact)

AI Doesn’t Fail. Services Do - From AI Use Cases to AI-Native Services and Sustainable Business Value

AI is no longer just a technology or a feature to add to an existing product. The real challenge is designing the services, organisations and business models around it.

Drawing on real-world Service & Business Design cases, we explore what happens when AI moves from a promising use case into a real service — and why many AI initiatives struggle not because the technology fails, but because the surrounding system was never designed for it.

We will look at how organisations can move from: Service & Business Design, AI Opportunity, AI-Native Operating Model, Adoption, Sustainable Business Value

Through practical cases, we will explore the new roles, decision flows, human–AI handovers, trust, accountability, governance and operational changes required when AI becomes part of the service architecture.

The key question shifts from: “Where can we use AI?” to: “What system are we designing — and where can AI create meaningful, sustainable value?”

Desislava Sarbinovska

Co-founder at Service Design Digital

Desislava Sarbinovska is the co-founder of Service Design Digital, an international agency specializing in service design, digital transformation, and responsible AI. She works with global organizations to redesign services, transform business models, and turn emerging technologies into meaningful business and social impact.

With 15 years of experience in the IT industry and a strong background in financial services, education, and digital innovation, Desislava combines strategic leadership with deep design expertise. She holds an Executive MBA from IEDC-Bled School of Management (Slovenia) and a specialization in Service Design for Innovation from IED Design School (Barcelona).

12:10-12:40

Bojan Klasan (A1 Hrvatska)

From Training to Inference: Redesigning the Data Center for the AI Era

Bojan Klasan

Head of ICT Delivery and Data Center Services at A1 Croatia

Bojan Klasan is the Head of ICT Delivery and Data Center Services at A1 Croatia, driving strategic planning, infrastructure design, governance, and end-to-end technical execution across data centers, cloud environments, and complex enterprise ICT solutions.

With over 20 years of industry experience, he specializes in bridging technical infrastructure with business strategy. His leadership encompasses full-lifecycle data center projects—from initial design and budgeting to deployment and optimization—alongside the seamless delivery of high-availability ICT services, cloud integration, and enterprise-grade architecture.

12:40-13:30

PANEL by CroAI

Your AI agent has access to everything. What could go wrong?

Moderator: Tomislav Vazdar (Riskoria Advising & Professional Services)
Panelists: Marko Gulan (Astera Advisory), Bojan Ždrnja (Infigo IS)

What happens when AI agents move beyond the role of assistants and become systems that can independently access emails, documents, databases, code and other business tools, and take action? This panel explores the security of AI agents and models, access and permission management, and secure AI development, alongside the broader implications for cybersecurity, risk and governance. The discussion also examines human oversight and accountability: how do we maintain control, and who is responsible when an agent makes a mistake or is granted excessive permissions?

Tomislav Vazdar

Founder and CEO at Riskoria Advising & Professional Services

Tomislav Vazdar is the Founder and CEO of Riskoria Advising & Professional Services, an advisory firm working in cybersecurity, risk management and regulatory compliance. His career spans international roles in corporate IT leadership and risk management advisory, followed by more than a decade leading corporate security at one of Croatia's largest banks.

He advises essential and important entities on implementing the Croatian Cybersecurity Act and the NIS2 Directive, with a focus on risk management and the resilience of the systems that health, finance, energy, supply and other essential services depend on. He also works on AI governance and the threats posed by autonomous AI systems.

He serves as Programme Lead for the MSc Enterprise Cybersecurity at OPIT - Open Institute of Technology and as a Board Member of the Croatian Artificial Intelligence Association (CroAI). As a member of the NKS community, he is an active stakeholder in Croatia's cybersecurity ecosystem. He holds a degree in finance from the Faculty of Economics and Business, University of Zagreb, where he is completing his doctoral studies in Economics and Global Security.

Marko Gulan

CEO at Astera Advisory

Marko Gulan is CEO of Astera Advisory and an advisor to boards on cybersecurity, risk management and corporate governance. With more than 15 years in the ICT sector, his career began at Microsoft, continued in the telecommunications industry, and led him to Schneider Electric, where he advised on critical infrastructure and OT/ICS security across South-East Europe. Today, as an externalized CISO, he guides organizations across critical infrastructure, financial services and industry through compliance with the NIS2 Directive, DORA, the Cyber Resilience Act (CRA) and ISO/IEC 27001, and is a certified ISO/IEC 27001:2022 Lead Auditor.

His current work centres on risk and compliance management in the adoption of artificial intelligence, and on what happens when organizations deploy AI faster than they define accountability for it. He is a regular speaker at domestic and regional conferences and a columnist for leading business and technology publications. He also shares his experience with students as a lecturer in professional cybersecurity courses, with a growing focus on the risks introduced by AI.

Bojan Ždrnja

CTO at Infigo IS

Bojan Ždrnja is the CTO of Infigo IS (part of the Allurity Group) and a cybersecurity expert with extensive experience in penetration testing, red teaming and security service development. As a SANS Principal Instructor and co-author of the SEC542 course, he trains cybersecurity professionals and explores emerging technologies and attack techniques. In recent years, his work has focused on the use of artificial intelligence in offensive security and how AI is reshaping the speed, scalability and sophistication of modern cyberattacks.

13:30-14:30

LUNCH

14:30-15:00

Stefan Martinić (Attorney at Law) & Kristijan Galić (Lawyer at Galić i Martinković)

Croatia Made (Certain Uses of) AI a Crime. Are You Next?

Croatia has done what no other country dared to do: it wrote certain artificial intelligence uses directly into its Criminal Code. Is this a bold step toward safer, more responsible AI, or a chilling effect that will send innovation, talent and capital across the border? Engineers, founders, employers, investors, IT professionals, lawyers, architects: is every person in this conference one prompt away from a criminal complaint, or will the new offence never leave the statute book and remain a dead letter? Two attorneys who deal with these questions daily go head to head. Kristijan Galić defends the law. Stefan Martinić argues it should never have been written. Twenty minutes, one on one, no moderator to hide behind. You decide who is right

Stefan Martinić

Attorney at Law, LL.M. Harvard Law School

Stefan Martinić is an attorney with years of experience in leading Zagreb law firms, specializing in corporate law, European Union law, and technology law (including GDPR, artificial intelligence, intellectual property, etc.). He has held all positions—from trainee to partner—before establishing his own law office.

He gained his education and experience at some of the world’s leading institutions: he earned an LL.M. from the London School of Economics (EU Law & Tech Law) and Harvard Law School (Cyberlaw), participated in an exchange program at Université Paris Panthéon-Sorbonne, obtained the CIPP/E certification and an AI certification from Saïd Business School (Oxford), and completed internships at the Court of Justice of the EU and the European Parliament.

He is a member of the working group of the Ministry of Justice, Public Administration and Digital Transformation of the Republic of Croatia for drafting the national law implementing the AI Act.

Kristijan Galić

Lawyer at Galić i Martinković j.t.d.

Attorney at law Kristijan Galić finished his law studies at the University of Zagreb in 2017, after which he has been employed as a trainee lawyer, acquiring extensive experience across almost all legal branches, excelling particularly in the criminal law, constitutional law and civil law.

In 2021, he passes the bar exam before the Ministry of Justice and Administration of the Republic of Croatia, after which has been registered as the Attorney at law in the Croatian Bar Association. In 2022, he passes the bankruptcy trustee exam before the Ministry of Justice and Administration of the Republic of Croatia and obtains a certificate for the public procurement specialist before the Ministry of Economy and Sustainable Development of the Republic of Croatia.

In 2023, together with Attorney at law Marko Martinković, he is founding the law firm Galić i Martinković j.t.d., continuing to provide comprehensive legal services to both domestic and foreign clients.

15:00-15:30

Vedran Jelić (A1 Hrvatska)

Press 1 for AI

Vedran Jelić

Digital and AI Specialist in Customer Service at A1 Hrvatska

Vedran Jelić is a Digital and AI Specialist in Customer Service at A1 Hrvatska. An educator by training, he earned his master's degree in education (mag. educ.) from the Faculty of Croatian Studies, University of Zagreb, in 2021. and started his career at A1 in 2022, in customer service. Today he works on bringing digital and AI solutions to both customers and frontline teams. An AI enthusiast at heart, Vedran loves running trainings and helping people learn, cares deeply about customers, and is excited by everything AI makes possible.

15:30-16:00

Marko Martinović (Margins Agency)

From Pilot to Production: Building an AI System for Field Sales Teams

In wholesale distribution, one sales rep often covers 500+ customers alone. The signals that should decide which customer to visit next come from at least six disconnected places: management emails, the promotion calendar, seasonality, field escalations, competitor information and collection reports. In practice, visits are planned from memory and habit.

This talk shows how we built an AI system for the field sales team of a regional B2B distributor with thousands of active customers. The system has three parts. The first is an ML model that learns the buying rhythm of each customer separately and warns the rep about significant deviations, instead of using one fixed threshold for all. The second is visit planning with AI suggestions based on five weighted signals, together with voice reports that are transcribed on the device and structured by AI. The third is a conversational assistant that answers questions about the rep's own customers, with record-level access control and the data source and period stated for every number.

We will focus on the decisions that turned out to be more important than the model: keeping all business data inside the client's environment, respecting existing master data processes instead of bypassing them, GPS tracking without turning it into surveillance, and giving reps flexibility, because a strict calendar would be rejected. We will also share first experiences from the rollout to the sales team.

Marko Martinović

Partner and Regional Director at Margins Agency

Marko Martinović is Partner and Regional Director at Margins, a Zagreb-based company that has delivered complex software and AI solutions for mid-size and large companies, mainly in the United States and Australia, for more than eight years. In 2025, Deloitte recognised Margins as the fastest-growing tech company in Croatia and the 11th fastest-growing tech company in Central Europe. Based on this experience, Margins built Rivermind, an AI platform for enterprise operations. Marko leads Rivermind's work with enterprise clients in Croatia and the wider region, where the team implements AI systems in production across distribution, manufacturing, retail, financial services and other industries.

11:40-12:10

Luka Kladarić - TBC

12:10-12:40

Gordan Kreković & Hrvoje Šamija (Visage Technologies)

Gordan Kreković

CEO at Visage Technologies

Gordan Kreković is the CEO of Visage Technologies, a company providing R&D services and products based on edge AI and computer vision. With a PhD in Computing and a diverse career managing tech products, teams, and companies, Gordan has broad experience spanning research, engineering, and management. Alongside his main career, he has authored and co-authored numerous scientific publications, served as an Assistant Professor at the Academy of Music, remained active in the FER alumni community by serving as president of AMAC-Alumni FER, and continued his engagement in music and composition.

Hrvoje Šamija

CBDO at Visage Technologies

Hrvoje Šamija is a technology executive and engineer who serves as the Chief Business Development Officer (CBDO) at Visage Technologies, a company specializing in computer vision, artificial intelligence, and machine learning solutions. Alongside his team, Hrvoje focuses on bridging the gap between deep-tech research and market-ready industrial applications, driving the commercialization and development of advanced solutions spanning robotics, automated driving, and edge AI. He holds a Master’s degree from the Faculty of Electrical Engineering and Computing (FER) at the University of Zagreb, as well as an MBA from COTRUGLI Business School. Prior to his current position, Hrvoje built a strong foundation in software and hardware development, serving in various engineering and leadership roles.

12:40-13:10

TDSynnex & NVIDIA

13:10-13:30

Juraj Pejnović (EPAM Systems)

Your AI Agents Need a World Model Using Ontology for Context, Tool Safety and Persistent State

LLM agents can reason over language and call increasingly powerful tools, yet most have no explicit model of the systems they operate in. The domain model is already present — fragmented across application types, database schemas, APIs, Jira, Git, documentation and business rules — but agents reconstruct it repeatedly from text.

Through a controlled software-engineering experiment, we compare the same task with and without ontology guidance and examine where the approach improves reliability, context efficiency, tool safety and traceability — and where the additional modelling cost is not justified.

Juraj Pejnović

Senior software engineer at EPAM

Juraj is a senior software engineer in EPAM with more than five years of experience building large-scale web applications, primarily with React and TypeScript. Alongside engineering, he is interested in AI-assisted development, agentic workflows, software architecture, and the way new tools change how engineering teams organize their work. He also runs his own e-commerce business and is currently pursuing an MBA at ZŠEM in Zagreb.

13:30-14:30

LUNCH

14:30-15:00

Ivan Projić (Neuromorphyx)

15:00-15:30

Ivan Vican (Muse Group)

Listening, Composing, Notating: Lessons from Building AI for Music

Ivan Vican

Staff Machine Learning Engineer at Muse Group

Ivan Vican is a Staff Machine Learning Engineer at Muse Group, the company behind MuseScore, Audacity and Ultimate Guitar, where he develops advanced audio models for millions of musicians, most recently an audio-to-score system now in production in MuseScore. Before that he spent three years at Aiva Technologies, one of the earliest companies working on generative music, where he developed a best-in-class neural audio codec that became the foundation of their music generation models. Over the course of his career he has consulted for Fortune 500 companies, led AI teams, and developed dozens of deployed machine learning and signal processing solutions across music, healthcare, agriculture and consumer electronics. Ivan holds a PhD in audio machine learning, is a co-inventor on six patents in audio technology, and has authored scientific publications on audio, acoustics and biomedical signal processing.

15:30-16:00

Mihael Španović (S.A.G.E.)

The Prompt Works. Now What?

Most teams build an AI feature the same way: write a prompt and send every request to a large language model. For a prototype, that is a good choice. Once the feature runs thousands of times a day, many of those calls are routine decisions that a much smaller model could make faster and for less money. This talk covers how to hand those decisions to smaller, faster models, with the LLM helping to build them and staying on for the hard cases. The examples come from agricultural data pipelines and cover where exampls such as [JEV] and open-weight models fit in.

Mihael Španović

Lead Engineer at S.A.G.E.

Mihael Španović is Lead Engineer at S.A.G.E., a public benefit company building an AI-native agricultural intelligence platform spanning soil and crop modeling, carbon intensity scoring, and insurance products for US agriculture. He previously worked as Head of Engineering at Protostar Labs, leading a 10-person team across BioTech, industrial vision, and space sector projects. Mihael holds an M.Sc. in AI from Johannes Kepler Universität Linz and conducted robotics research at the University of New Mexico. Outside of work: motorcycles, airsoft, board games, and hardware tinkering.

Business & Implementation Track
Engineering Track
Research Track

10:00-10:30

Rinaldo Ugrina (Baymard Institute)

How AI Designs the The New Logic of Work

"AI won't take your job, but someone using AI will." You hear it everywhere, and it hands you just enough clarity to stop asking the harder question.

It tells you to look at your tasks and ask whether AI does them for you or helps you do them better. Meanwhile the thing actually being rearranged is the system those tasks sit inside. Which work gets done in what order, who owns which call, and what the organization is willing to pay for.

Tasks rarely lose value because the people performing them got worse. They lose value because they stop creating an advantage. So the question worth asking is not how do I do my work better with AI. It is which constraint was my work removing, whether that constraint is still there, and which one took its place.

The talk covers:

Why automate versus augment is the wrong unit of analysis, and what to look at instead

How to find the constraint your work actually clears, and whether AI has already removed it

Explains why your organization's AI investment is not showing up as faster decisions

Rinaldo Ugrina

Head of Product at Baymard Institute

Rinaldo Ugrina is Head of Product at Baymard Institute, a UX research firm whose research is used by 71% of all Fortune 500 e-commerce companies. He has run over 200 high-stakes sessions with product and design teams at companies such as Google, Amazon, Meta, Stripe and Nike, which gives him an unusually direct view of how large organizations actually make decisions. He is a keynote speaker on how ideas move inside organizations, securing organizational buy-in and on where value goes when AI makes execution cheap.

10:30-11:00

TBD (Sedmi odjel)

11:00-11:30

COFFEE BREAK

11:30-12:00

Chris Thomas (LSE Data Science Institute)

AI’s impacts on the quality of jobs

This presentation will explore the implications of AI adoption for the conditions and experience of work. Drawing on recent LSE research, it will consider relevant dimensions of job quality, examining existing evidence on how AI may be affecting them, and identify the most important open questions and priorities for workers, organisations, and governments.

Chris Thomas

Policy Fellow at the LSE Data Science Institute

Chris Thomas is a Policy Fellow in the LSE Data Science Institute. His work focuses on the development of AI governance mechanisms and regulatory policy to support ethical AI and the management of AI-driven societal transformation.

At LSE, Chris leads on work supported by the Foreign Commonwealth and Development Office advising policymakers around the world on AI regulatory strategy and capability development. Chris also plays a central role in delivering the LSE Global Forum on AI and the Social Sciences and producing its accompanying State of AI and Society report.

Before joining LSE, Chris was a Senior Research Associate at The Alan Turing Institute where he led a project on regulatory readiness for AI and advised UK regulators on emerging policy challenges. He also led research for the AI Standards Hub, examining the role of standards in the governance of foundation models.

Alongside his role at the Turing, Chris served as a Senior Policy Advisor to the UK Department for Science, Innovation and Technology, via the Government’s Expert Exchange Programme. Chris was the lead author of the AI Regulatory Capability Framework and Self‑Assessment Tool, co-developed with UK regulators and government departments.

Chris co‑authored the AI Assurance Roadmap at the Centre for Data Ethics and Innovation and has contributed to the UK Government’s AI Strategy (2021) and AI Policy Paper (2022). He has represented the UK as an expert in the Council of Europe’s CAHAI Policy Development Group, contributing to the Human Rights, Democracy and Rule of Law Impact Assessment for AI.

Previous positions include at the Oxford Internet Institute, where Chris co-authored a methodology for interpreting and applying ethical AI principles in high‑risk settings, and at the Ada Lovelace Institute, where he researched the ethical and societal implications of health data use outside of clinical settings.

12:00-12:30

Toma Petrač (Span)

From Chat to Platform: Building a Multi-Agent AI Workspace for Enterprise Knowledge

We will present how the project evolved from a conversational interface into a multi-agent platform designed for enterprise knowledge work. The talk will cover the agent management layer, coordinator-driven orchestration, human-in-the-loop control, fast RAG over structured and unstructured sources, and a document reasoner agent that can process multiple documents in a single flow. We will also show how workspace-level controls let customers manage sources, permissions, prompts, language, default models, and which agents are available per workspace. A separate evaluation platform supports regression testing across corpora, precision, faithfulness, key fact coverage, and time-to-first-token so changes can be validated safely over time. Finally, we will explain how the architecture is being shaped for future interoperability with Copilot Studio, Microsoft Copilot agents, and hybrid or on-prem deployments.

Toma Petrač

Machine Learning Specialist at Span

Toma Petrač is a Machine Learning Specialist with experience designing production-oriented AI, machine learning, and document intelligence solutions. He is currently focused on building a custom multi-agent copilot platform that combines retrieval-augmented generation, analytics, document reasoning, tool orchestration, and enterprise Azure services.

His work spans agentic architectures, prompt and context engineering, NLP, MLOps, and backend development, with a particular interest in making complex AI systems reliable, scalable, and practical for enterprise use. He also explores document intelligence and hybrid or on-prem deployment approaches for AI-powered knowledge systems.

12:30-13:00

Nenad Mandić (Abysalto)

Humanoids Beyond the Demo Reel: Teaching a Robot a Real Industrial Task

Language models can write code and pass professional exams, yet robots still struggle with tasks a child does without thinking. One reason is that today's AI learns from a dimensionally impoverished world. Text and images are a compressed shadow of physical reality, while a robot has to deal with mass, friction, contact, time and real consequences.

This talk uses that idea to look at a first year of building Physical AI at Abysalto. It covers setting up an embodied AI team and an NVIDIA Isaac simulation stack, and teaching a humanoid robot a real industrial task. It also includes early work with quadrupeds, vision systems and AR glasses. Expect an honest view of what is hard, what is realistic today, and what is still hype.

Nenad Mandić

Head of AI at Abysalto

Nenad is Head of AI at Abysalto, responsible for shaping and executing AI solutions and physical AI (robotics) strategy while building the capabilities for effective adoption of AI innovation in business.

With more than 20 years of experience in banking, payment systems and digital transformation, he has a proven track record in delivering solutions where reliability, execution and business value are critical.

His current focus is on developing AI solutions including LLMs, NLP, RAG, agentic systems, automation, and robotics, with strong emphasis on bringing AI solutions into real-world applications.

13:00-14:00

LUNCH

14:00-14:30

Hani Zahirović (Bloomteq)

Scaling and monitoring AI assisted software development at Bloomteq

Hani Zahirović

CTO at Bloomteq

While having encountered a variety of technologies throughout his professional career, when asked to describe himself, Hani likes to state that he considers himself a problem solver. Hani has been working withsoftware development teams in various capacities for over 10 years, assisting them in planning, writing, and testing code, as well as organizing processes, nurturing team growth, educating team members, and much more. As a natural progression, Hani has played a key role in organizing the company's tech department as the CTO, establishing a roadmap for the technical direction we aim to pursue in the upcoming period.

14:30-15:00

Antonella Barišić Kulaš (Odonata Technologies)

From Robotics Research to the Warehouse Racks

Antonella Barišić Kulaš

Co-founder and CEO of Odonata Technologies

Antonella Barišić Kulaš, PhD, is a co-founder and CEO of Odonata Technologies, an inventory intelligence company transforming warehouse operations. Odonata provides real time visual proof of physical stock, identifying missing, misplaced, or unrecorded items against WMS records so logistics teams make decisions based on audit ready evidence.

With a research background in robotics and computer vision, Antonella spent over six years developing AI systems for autonomous robots before founding Odonata. She holds a summa cum laude PhD from UNIZG-FER, has led NATO and EU research projects, is a L'Oréal-UNESCO For Women in Science laureate, and was part of the team that won the MBZIRC 2024 international robotics challenge.

15:00-15:30

Imran Nino Eškić (Hyperbunker)

Imran Nino Eškić

Co-founder of HyperBUNKER

Imran Nino Eškić is an innovator, co-founder of HyperBUNKER, and an expert with nearly 30 years of hands-on experience in data recovery and security. He is the founder of InfoLAB, the first data recovery laboratory in this part of the EU, with more than 50,000 data recovery cases handled when all seemed lost.

However, with the rise of AI technologies, sophisticated ransomware, and cyberattacks, technical data recovery is not always possible.

To ensure business continuity in the event of any attack, he invented HYPERBUNKER – a patented offline vault that is inaccessible to hackers and ransomware attacks. He holds a U.S. patent (US12608142B1) that sets new standards in cyber resilience. Today, he combines AI solutions with hardware-level security, focusing on the protection of critical infrastructure and shaping a secure technological future.

09:30-10:00

Dušan Omerčević (*codeplain)

AI code will be regenerated, not maintained

Dušan Omerčević

CEO and founder of *codeplain

Dušan Omerčević is CEO and founder of *codeplain, an AI-powered code generation platform that turns production-ready software from specifications written in ***plain language. Previously, Dušan founded Cleanshelf, a SaaS management platform, which was acquired by SAP LeanIX in 2021. He holds an M.Sc. in Computer Vision from the University of Ljubljana.

10:00-10:30

Michał Antropik (Neuromorphicism) - TBC

10:30-11:00

Marko Velić (Google)

Neuromorphic Computing and the new wave of AI

LLMs are very capable and useful systems (if used right), but their success has narrowed our picture of what artificial intelligence could be. A model trained by massive-scale backpropagation (in a huge data center) and then largely frozen at deployment is very different from a nervous system that continuously processes temporal signals, adapts locally, and operates under severe energy constraints. In the shadows of the ongoing focus on LLMs, there is an alternative, much more biologically plausible area of AI research.

This talk explores the intersection of computational neuroscience, spiking neural networks, and neuromorphic computing, an alternative computing paradigm. We will build a spiking neuron from first principles, examine how networks compute through temporal dynamics, and explore biologically inspired learning mechanisms and discuss remaining challenges in the field.

Marko Velić

Staff Software Engineer at Google

Marko is a Staff Software Engineer at Google where he work on Gemini model improvements for multimodal and educational usecases. Previously he worked at Photomath, Meta (Facebook), Styria, LEGO (as consultant), University of Zagreb and has entrepreneurship experience. He won many awards including the ones from Microsoft and NVIDIA. Marko holds PhD and MBA degrees. Marko is also a lecturer at Algebra University. He is interested in ML research and practical applications in real life scenarios.

11:00-11:30

COFFEE BREAK

11:30-12:00

Nedim Muhamedagić (Bloomteq)

AI at the Edge: Bridging Machine Learning and Embedded Systems

Nedim Muhamedagić

Chief Architect Officer at Bloomteq

With over a decade of software development experience, Nedim has taken on the Chief Architect Officer role at Bloomteq. He excels in navigating complex challenges with rapid problem-solving skills. Renowned for simplifying complexities, Nedim leads by example with a dedication to excellence. His specialty is optimizing dev team delivery at scale, ensuring streamlined processes and efficient outcomes.

12:00-12:30

Ivan Nikolov (Celtra)

From Prediction to Explanation: Building Creative Intelligence

Advertising platforms generate huge amounts of performance data, and a natural first step is to use that data to predict how well the creative asset used in an ad is likely to perform before it goes live.

At Celtra, we explored this through creative scoring: using machine learning to assign creative assets a predicted performance score before a campaign runs.

But while scoring can help identify stronger and weaker creatives, it does not necessarily explain which creative characteristics drive performance or what a marketer should change.

In this talk, we’ll walk through our evolution from pre-flight creative scoring to a more interpretable Creative Intelligence system. We’ll cover the challenges of modeling noisy real-world advertising data, why black-box scoring proved difficult to operationalize, and how we redesigned the problem around interpretability and actionability.

Our current approach uses multimodal large language models to extract interpretable creative attributes. We then use Bayesian modeling to estimate how those attributes relate to advertising performance while accounting for uncertainty. This lets us generate insights such as whether prominently featuring a product, including a person, or using a particular visual style is associated with better or worse performance.

The result is a system that moves beyond prediction toward insights marketers can understand and act on.

Ivan Nikolov

Machine Learning Engineer at Celtra

Ivan Nikolov is a Machine Learning Engineer working in Celtra. At Celtra, he has worked on the Creative Intelligence solution, focusing on operationalizing ML systems, Bayesian modeling, and AI engineering with frontier foundation models. His work spans unstructured image and video understanding, as well as transparent and explainable ML decision-making.

12:30-13:00

Zlatan Ajanović (Technikum Wien)

Zlatan Ajanović

Postdoctoral Researcher at the Technikum Wien

Dr. Zlatan Ajanović is a Postdoctoral Researcher at the Technikum Wien. Before, he was a postdoctoral researcher at TU Delft with Prof. Jens Kober. He earned his PhD degree from the Graz University of Technology. His main research interest lies in the Decision-making and Control for Autonomous Robots based on Planning, Learning, and Control Theory. He is the recipient of the IFAC Young Author Award, Hans List Award, and DAAD AInet PostDoc Fellowship. He has co-organized several workshops and conferences, the most notably AAAI25 and ICAPS25 Bridging the Gap Between AI Planning and Reinforcement Learning (PRL), ICAPS25 Workshop PlanRob: Planning and Robotics, ICRA 2023 Workshop on Life-Long Learning with Human Help (L3H2), 2023 BeNeLuX AI Conference, and ICAPS 2024 Conference.

13:00-13:30

Nikolina Kosanović (Njuškalo)

Category Classification, Reengineered: Fusing Text, Image, and Hierarchy

At a time when everyone is rushing to implement AI agents in their products, some are realizing they come with a cost - and not every problem needs one. We are happy we still have use cases that are best solved with a "good old" classifier which requires data crunching, model training, and most importantly - investigation.

Njuškalo has almost 2,000 ad categories, wildly imbalanced, arranged in a tree. This talk walks through the multimodal, hierarchical model rebuild - the challenges we faced with the data, the design decisions, and what we learned.

Nikolina Kosanović

Senior ML Engineer at Njuškalo

Nikolina Kosanović is a Senior ML Engineer with over 5 years of experience in Machine Learning. She works on the Data Science team at Njuškalo, Croatia's leading classifieds marketplace. Classification is a recurring thread in her work: recently, she has focused on multimodal classification models and agentic systems, and she was part of the team that won second place in the AI4Health 2025 competition, working on BI-RADS categorization of mammography images. At AI2Future, she will share how she reengineered category classification by fusing text, image, and hierarchy.

13:30-14:30

LUNCH

14:30-15:00

Eldar Kurtić (Red Hat)

Beginner-Friendly Introduction to Speculative Decoding: From Zero to Hero

A beginner-friendly introduction to speculative decoding that builds the idea from first principles and gradually connects it to modern LLM inference systems. We will cover why autoregressive generation is slow, the core draft-and-verify intuition behind speculative decoding, how it can improve latency without changing model outputs, and the main design choices that determine whether it works well in practice. The goal is to take the audience from having no prior knowledge of speculative decoding to understanding the key concepts, terminology, trade-offs, and practical considerations behind real-world implementations.

Eldar Kurtić

Principal Research Scientist at Red Hat and Institute of Science and Technology Austria (ISTA)

Eldar Kurtić is a Principal Research Scientist at Red Hat and Institute of Science and Technology Austria (ISTA), specializing in efficient inference techniques for large language models (LLMs), with a particular focus on pruning, quantization, and speculative decoding. His work centers on developing methods to accelerate inference within the vLLM engine, bridging cutting-edge research with practical deployment solutions.

15:00-15:30

Vlatko Kosturjak (Marlink)

Vlatko Kosturjak

VP of Research at Marlink Cyber

Vlatko Kosturjak is VP of Research at Marlink Cyber, with over two decades of experience in information security and cybersecurity. His varied roles have given him both a comprehensive understanding of security governance and deep technical expertise.

He has discovered numerous vulnerabilities in software and hardware, ranging from widely used DNS servers such as ISC BIND and Unbound to the Linux kernel. He also holds several professional certifications, including CISSP, CISA, CISM, and OSCP.

Across many years of commercial work in the field, he has gained successful M&A experience in various areas of cybersecurity and in different roles.

10:00-10:30

Stjepan Picek (FER)

AI Under Attack: Security in the Age of Intelligent Systems

Artificial intelligence is rapidly changing cybersecurity - but not only by helping us defend systems. AI is increasingly being used to discover vulnerabilities, automate attacks, generate malicious content, and coordinate complex cyber operations. At the same time, AI systems themselves introduce an entirely new attack surface.

Modern machine learning systems can be manipulated in ways that traditional software cannot. Attackers can alter their behavior through carefully crafted inputs, implant hidden backdoors during training, extract information from models, bypass safety mechanisms, or manipulate the internal components responsible for seemingly safe behavior. As AI systems become more autonomous and gain access to tools, data, and other systems, the consequences of such attacks become increasingly significant.

In this talk, I will explore this emerging security landscape through concrete and accessible examples, ranging from adversarial attacks and backdoors to jailbreaks and attacks on large language models and AI agents. We will look at AI from both sides of the security equation: as a powerful new tool for attackers and as a complex system that itself needs protection.

The central question is simple: as we give AI systems more capabilities, autonomy, and responsibility, are we learning how to secure them quickly enough?

Stjepan Picek

Full professor at the University of Zagreb Faculty of Electrical Engineering and Computing

Stjepan Picek is a full professor at the University of Zagreb Faculty of Electrical Engineering and Computing, an associate professor at Radboud University, and an adjunct professor at the University of Bergen. His research spans security and cryptography, machine learning, and evolutionary computation, with particular interests in security of machine learning.

He earned a PhD (double doctorate) in computer science in 2015 from the University of Zagreb and Radboud University, and a PhD in mathematics in 2024 from the University of Paris 8. Before his current appointments, he held positions at TU Delft, MIT, and KU Leuven, alongside experience in industry and government.

Professor Picek has delivered more than 70 invited talks and published over 200 refereed papers. His work has been recognized with several awards, including the Vera Johanides Award, the Rikard Podhorsky Award, and the IEEE Croatia Section Award. He is a Senior Member of IEEE, a member of ELLIS, a Fellow of the Young Academy of Europe, and serves as an associate editor for several international journals.

10:30-11:00

Prof. dr Elmir Babović (Blum institut)

AI at the Edge: Bridging Machine Learning and Embedded Systems

Elmir Babović

Research Director at the Blum Institute

Dr. Elmir Babović is a professor at the Faculty of Information Technology (FIT) in Mostar and the Research Director at the Blum Institute, the first AI research institute in Bosnia and Herzegovina. With a strong background in applied AI, robotics, and industrial automation, he leads research initiatives that connect academic development with real-world industry applications, focusing on energy, automated systems, and practical AI implementation.

11:00-11:30

COFFEE BREAK

11:30-12:00

David Puljiz (NEURA Robotics)

From Tokens to Torques: Physical AI and the Humanoid Moment

AI is no longer just software on a screen —it's starting to move through the real world, and that changes everything. This talk explores the rise of physical AI: intelligence that has to reckon with gravity, friction, and time, not just data. We'll look at why humanoid robots are having their moment right now, why they need AI to function, and how AI and hardware are increasingly shaping each — every advance in one exposing new limits, and new possibilities, in the other. Along the way, we'll unpack what's actually holding this technology back, and what it will take to close the gap between thinking machines and moving ones.

David Puljiz

Researcher and project coordinator at Neura Robotics

David Puljiz is a researcher and project coordinator at Neura Robotics, a German company building Physical AI and cognitive robots, such as the 4NE1 humanoid. His work focuses on easing the programming of and interaction with robots — spanning robotics, human-machine interaction, mixed reality, and AI — with the goal of making robots easier to deploy and genuinely useful in the real world.

A graduate of the Faculty of Electrical Engineering and Computing at the University of Zagreb, David went on to work as a research assistant at the Karlsruhe Institute of Technology (KIT), following an internship at Robert Bosch GmbH in machine vision and image processing. He obtained his PhD from KIT in 2024 for his work on using augmented reality to make industrial robots more intuitive to use. At AI2Future, he will share how AI is reshaping robotics and the way humans and robots work together.

12:00-12:30

Jelena Lončar & Filip Mirković (Atomic Intelligence)

Fake it Till You Make it: Generating (and grading) synthetic tables

Can one model learn to generate high-quality synthetic data across many different tables? And what does "high-quality synthetic data" even mean? This talk explores the promise of tabular foundation models and presents our approach in the CopyDat–SyntDataHub project. We’ll take an accessible technical look at what it takes to build a generator that can adapt to tables with different schemas, data types, and domains. We’ll also examine what makes synthetic data good, and why grading it can be just as challenging as generating it.

Jelena Lončar

Machine Learning Research Engineer at Atomic Intelligence

Jelena Lončar has more than five years of experience in the research and development of innovative AI-based solutions. She is currently a Machine Learning Research Engineer at Atomic Intelligence, where her work on the CopyDat–Synthetic Data project focuses on novel methods for generating high-quality synthetic tabular data from sensitive and complex datasets. The project aims to enable safer data use and sharing while preserving the original data's statistical and analytical value. Previously, she worked as a consultant on ADAS systems, developing and optimizing production Computer Vision algorithms for the real-time detection and tracking of vulnerable road users.

Filip Mirković

Machine Learning Research Engineer at Atomic Intelligence

Filip Mirković has more than three years of experience in the research, development, and deployment of production-grade deep learning systems. He is currently a Machine Learning Research Engineer at Atomic Intelligence, where he works on the CopyDat–Synthetic Data project, developing novel methods for generating high-quality synthetic tabular data from sensitive and complex datasets. The project aims to enable safer data use and sharing while preserving the statistical and analytical value of the original data, combining prior-data fitted networks (PFNs), variational autoencoders, and flow matching with methods for quantifying privacy risk. His work also extends to AI for music and biomedicine, including deep learning models for audio source separation and predictive models fusing peptidomic biomarkers with prostate MRI segmentation to support prostate cancer detection. He holds an MSc in Physics from the University of Zagreb, with a thesis on symmetry-aware graph neural network frameworks for modeling complex systems.

12:30-13:00

Paul Duncan (UK’s National Physical Laboratory (NPL))

Paul Duncan

Principal Scientist at the UK’s National Physical Laboratory (NPL)

Dr Paul Duncan is a Principal Scientist at the UK’s National Physical Laboratory (NPL), where he leads the Informatics group within the Data Science & AI department. With a background in data modelling and uncertainty evaluation, his work focuses on developing data quality frameworks and methodologies for understanding uncertainty in complex, interconnected systems. Dr Duncan is the Technical Lead (2022-present) for NPL’s involvement in the AI Standards Hub, where he plays a key role in shaping national and international approaches to Trustworthy AI and related standards. Dr Duncan is currently the co-chair of the BIPM Task Group on Data Quality & FAIR Practices in Metrology which aims to standardise approaches to Data quality across the world of metrology. His current research explores how metrology can underpin the development of reliable, transparent, and accountable AI systems. He is actively involved in cross-disciplinary projects that span data science, semantic technologies, and AI governance, and frequently collaborates with government, academia, and industry to advance confidence in the use of AI.

13:00-14:00

LUNCH

14:00-14:30

Barbara Arbanas Ferreira (MARBLE)

14:30-15:00

Jonatan Lerga (RI TEch)

AI Across Engineering Disciplines: A Snapshot of Research at the University of Rijeka, Faculty of Engineering

Artificial intelligence is increasingly reshaping how engineering challenges are approached and solved. This talk presents recent work by the research group at the Department of Computer Engineering, Faculty of Engineering, University of Rijeka, focusing on methods developed over the past several years that combine digital signal-processing algorithms with machine-learning techniques. Particular attention will be given to their applications in medical image and signal analysis, underwater image processing, remote sensing, seismology, hydrology, and maritime engineering

Jonatan Lerga

Full Professor of Electrical Engineering and Associate Dean for Education at the University of Rijeka, Croatia

Prof. Jonatan Lerga is Full Professor of Electrical Engineering and Associate Dean for Education at the University of Rijeka, Croatia. He received his PhD from the University of Zagreb in 2011. He founded and directs the University of Rijeka Center for Artificial Intelligence and Cybersecurity and heads the Information Processing Laboratory. Prof. Lerga has authored more than 110 scientific papers, including 30 Q1-ranked journal articles, and has led numerous EU Horizon, Digital Europe, and national research projects. His work has been cited over 1,100 times (Google Scholar). Honors include the 2022 Croatian State Science Award, the Croatian Academy of Sciences and Arts Award, and several regional and university prizes. He serves on editorial boards of international journals, reviews extensively, and is a member of UNESCO’s International Research Centre on AI and Croatia’s National Council for Digital Transformation. His research focuses on digital signal processing, artificial intelligence applications, coding, and cryptography.

"AI2Future is conference made by community for community. Since 2017, we are going beyond the hype and aiming to cover practical, real-world use cases interesting for engineers and researchers as well as practitioners from other industries. This year’s lineup of speakers will be the strongest so far, so make sure not to miss it!"

- Marko Velić (Software Engineer @ Google)

"10 years of the AI2Future conference have made our AI/ML community stronger and relevant on the EU stage. AI2Future is the place to learn, meet AI/ML peers from Croatian and regional companies, and from the whole CroAI community."

- Davor Aničić (CEO and Co-founder @ Velebit AI, Board Member @ CroAI)

"AI2Future is a great place to learn and connect with experts in the field. The conference provides just right mix of deep technical talks, visionary keynotes, and honest discussions about trends and challenges. Can wait for the next one!"

- Domagoj Kovač (Business Development Manager @ Sedmi odjel)

MEET THE SPEAKERS

AGENDA

PARTNERS 2025

Top

Platinum

Gold

Exibitors

Community Partners

About AI2Future

Venue

DATE & TIME

Gallery

Program Committee

ABOUT THE ORGANIZER

SUPPORTED BY CroAI

GET YOUR TICKET

We are sending your data!

Thanks for subscribing!

Ooooops!

CONTACT

INFO