LIST and the Luxembourg Competition Authority launch phase two of their AI partnership to better analyse ever-growing volumes of data.
Started in September 2024, the collaboration between the two public institutions enters a more ambitious phase: by August 2027, a single prototype will combine AI agents and visualisation to help case handlers, with decisions staying human.
Marcos Gomez, Research Engineer at LIST specialising in low-code and AI agents, looks back on phase one, reliability safeguards and what comes next.
Photo caption
From left to right: Daniele Pagani, Marco Estanqueiro, Jordi Cabot, Petar Georgioski, Marcos Gomez, Sofie Hancer, Alexandre Kraynyukov, Isabella Lorenzoni, Maël Cornil, Nicolas Médoc, Francesco Ferrero.
“[…] we are building on tools already used at European level […]”
Marcos Gomez, Research Engineer at the Luxembourg Institute of Science and Technology (LIST)
What did you learn from phase one that shaped the tech choices for this new phase?
The first phase, launched in September 2024, was above all a period of mutual learning. Our researchers gained a detailed understanding of how the Authority’s teams work and of the rigour and traceability a public enforcement body requires. The Authority’s teams got a clear view of what AI can and cannot realistically do today. That shaped our approach for this second phase. Rather than starting from scratch, we are building on tools already used at European level, and we are using BESSER, LIST’s own low-code platform, to accelerate development. We are also adding a visualisation component that helps make sense of the links within large volumes of information.
What kinds of documents or data will the AI agents handle first?
The focus is on the large, varied sets of documents that any investigation involves today, and evidence is increasingly digital. The first building blocks will help organise and classify incoming material, help teams search through and summarise long documents, and identify key information. In every case, the tool supports the reading and sorting work. The analysis and the legal assessment remain entirely with the Authority’s staff.
“Reliability is the central criterion of the project.”
Marcos Gomez, Research Engineer at the Luxembourg Institute of Science and Technology (LIST)
How will you ensure reliability and avoid bias in the AI models used?
Reliability is the central criterion of the project. We will compare several approaches, including large language models and more established language-processing techniques, and keep only those that prove most reliable and best suited to a public authority. Evaluation is part of every development cycle. The Authority’s teams test each new feature, and their feedback, together with measurable indicators, directly shapes the next cycle. So if an issue of reliability or bias appears, it is addressed straight away. The tool is also designed to assist, not to decide: its outputs are always reviewed by case handlers, and legal assessment and decision-making remain firmly in human hands. Finally, confidentiality is a prerequisite, so data will be processed in a secure, controlled environment that meets the requirements of a public authority.
What budget and team size are behind this phase through August 2027?
What makes this project work is less a question of headcount than of complementary expertise. On LIST’s side, researchers bring expertise in AI and language processing, to design and refine the different assistants, and in data visualisation, to turn complex information into clear, interactive maps. This close collaboration between domain experts and researchers keeps the tool grounded in practice.
“[…] making the tool available to other competition authorities has not been excluded […]”
Marcos Gomez, Research Engineer at the Luxembourg Institute of Science and Technology (LIST)
Any plans to share or roll out this prototype with other competition authorities in Europe?
The possibility of making the tool available to other competition authorities has not been excluded. The immediate goal, however, is a working integrated prototype for the Authority by August 2027.