Many companies are racing into AI without a clear strategy, resulting in pilots that don’t scale, fragmented tools, and rising risk exposure. The root causes are rarely technical; they stem from misaligned goals, immature data foundations, weak change management, and superficial governance.

This keynote reframes AI as an enterprise capability—anchored in business outcomes, powered by robust data and operating models, and governed by risk-by-design. Attendees will learn a practical playbook to move beyond demos: choose the right problems, prepare people and processes, architect integration, and institutionalize responsible AI practices so AI compounds value instead of compounding complexity.

Large Language Models (LLMs) have rapidly become the central drivers of modern AI, enabling unprecedented progress in reasoning, generation, multimodal understanding, and task automation. Their transformative impact reaches across sectors—from education and public services to cybersecurity, governance, and enterprise workflows.

In this keynote, I will discuss the foundations of LLMs, beginning with their core architecture based on the Transformer, the role of large-scale pretraining, and the methods used for fine-tuning, alignment, and safety, including supervised fine-tuning, LoRA/QLoRA, RLHF, and DPO.

I will also examine emerging trends such as agentic AI, model distillation, retrieval-augmented generation (RAG), and the shift toward efficient small and domain-specific LLMs. Special attention will be given to low-resource languages and the unique challenges and opportunities for developing localized LLMs for Azerbaijani and the broader Turkic language family.

Finally, real-world examples from academic and industrial projects will illustrate how LLMs can be deployed responsibly and effectively in national-scale applications—including speech technologies, digital government, security, and citizen services—highlighting key lessons and strategic directions for future development.

This keynote explores how neuro-symbolic AI can make robots more helpful, trustworthy, and intuitive for people to work with. Neuro-symbolic AI blends two powerful approaches: neural networks, which help robots see and recognise patterns in complex, changing environments, and symbolic reasoning, which gives them structured knowledge, logical thinking, and an understanding of human goals and context.

The talk shares examples from recent projects at Cardiff’s Research Centre in AI, Robotics and Human-Machine Systems (IROHMS). In these projects, neuro-symbolic AI helps robots better understand both the obvious and more subtle aspects of human behaviour, leading to smoother and safer collaboration.

A key benefit of this approach is that it makes robot decision-making more transparent, which can build trust and confidence. It also allows robots to learn from new data while still following clear rules, helping them stay safe, reliable, and aligned with human values—especially important in fields where safety and ethics matter.

Finally, the keynote discusses how neuro-symbolic AI could increase productivity and reshape roles across industry, education, and public services.

By creating digital humanities tools for researchers in history of science, the EIDA project – started in February 2023 – incorporates artificial intelligence into the study and analysis of a corpus of manuscripts of Ptolemaic tradition and early prints of mathematical astronomy. With the help of computer vision algorithms to accelerate specific steps (such as visual elements extraction, clustering, and vectorization) of the processing of the sources, EIDA opens new perspectives for the study of astral diagrams. The information system created for the project, based on a data model enabling the detailed description of the sources, will provide a diversity of tools dedicated to the exploration, analysis, and edition of diagrams in historical sources.

This presentation synthesizes how computer vision algorithms can be integrated to the work of historians and automate multiple steps of the processing of their source material, and how digital tools can be built in collaboration with researchers to both support their work and lay the groundwork for a public platform featuring reusable resources.

The development of these tools enable their application to various academic research projects currently underway at UFAZ. They prove invaluable in classifying and synthesising historical data stored on various media (written files, typed files, Word files, etc.), and processing images such as cartographic records to extract relevant information.

In recent years, new methodologies have allowed building agent-based simulation software executing on grid-shaped cell spaces. There have been numerous efforts integrating agents and cellular models for simulation, in which we can model agents that have emerging behavior that can represent intelligent systems. We have defined a new methodology for modeling such cell-based agent models using a formal modeling technique that permits defining each cell in a cell space as individual independent entity, called Cell-DEVS. The goal of Cell-DEVS is to build discrete-event cell spaces, improving their definition by making the timing specification more expressive and the definition of complex models simpler.

We will introduce the main characteristics of the Cell-DEVS formalism, and will show how to model complex cell spaces using this methodology. We will present different examples of application, and discuss open research issues in this area, focusing on models with emerging behaviour that can be used in AI applications.

We will then show some examples of the current use of DEVS, including applications in different fields. We will introduce an integrated environment that deals with these issues, orchestrating a cellular-based simulator (CD++), a GIS and data visualization, to simulate behavior and analyze results supporting the decision making for varied environmental scenarios. The limitations above are solved by adding raw simulation results into the georeferenced maps, associating many sources of information, providing a more powerful analysis experience.

The simulation model is fed by the GIS with updated data, while the model design process enables integrating additional information layers. The methodology uses a cellular modeling approach in which each cell is defined as a discrete event agent, and defines a procedure to couple cells evolving the state of the influenced neighbors. We will also discuss models of spiking neurons, data mining and market evolution, among others.

Digital transformation has accelerated the need for robust, interoperable, and scalable modelling and simulation frameworks capable of supporting complex socio-technical systems. From Industry 4.0 to smart cities, organizations increasingly rely on digital twins and distributed simulation to anticipate system behaviour, evaluate decision scenarios, and continuously optimize operations.

In this keynote, I will explore the methodological foundations and practical enablers of interoperable simulation, with a focus on DEVS formalism, model continuity, executable architectures, and standards such as HLA. Particular attention will be given to the integration of simulation with AI-driven components, enabling intelligent orchestration, predictive behaviour, and automated adaptation within cyber-physical ecosystems.

I will discuss recent advances in hybrid approaches combining agent-based modelling, discrete-event simulation, and data-driven methods to support real-time, collaborative, and human-in-the-loop applications. Examples from European and international research projects will illustrate how digital twins are transforming engineering practices, industrial transformation, healthcare system management, and large-scale decision-making.

Finally, I will highlight open challenges and strategic directions for the next generation of distributed simulation environments—particularly regarding trust, interoperability, sustainability, and the alignment of digital twins with organizational knowledge and governance requirements.