Founded in 2000 as a small team focused on preference modelling and fuzzy set theory, KERMIT (an acronym for “Knowledge Extraction, Representation and Management using Intelligent Techniques”) has grown into a leading research unit shaping the future of intelligent techniques and their applications. Over time, KERMIT evolved into a comprehensive team spanning all stages from data analysis to decision-making, with a focus on knowledge-based, predictive and spatio-temporal modelling paradigms. By maintaining a unique balance between theoretical advancements and practical applications, KERMIT has achieved remarkable success in output, visibility, and recognition. To accommodate growing specialization and enhance its reach, three subunits officially branched off in 2024: BionamiX, BioML and Biovism. Despite this structural evolution, KERMIT remains dedicated to its holistic philosophy, integrating diverse disciplines to tackle complex challenges.

KERMIT’s mission is to harness mathematics and computation to unravel life's complexities, optimize biological functions, and drive innovation in biodesign and decision-making under uncertainty. Focused on applied biological sciences—including biotechnology, environmental technology, plant breeding and food technology—, KERMIT refines existing methods and develops cutting-edge approaches across disciplines. The team is committed to creating accessible software tools that transform data streams into actionable and interpretable insights. Valuing continuous learning, interdisciplinary collaboration, and mental well-being, KERMIT embraces a holistic approach to solving challenges in our data-driven, interconnected world.

Mathematical modelling at KERMIT emphasizes intuitively appealing, rule-based paradigms—such as fuzzy modelling, cellular automata, and formal concept analysis—as well as cross-fertilizations thereof. The team has a particular interest in exploring the underutilized diversity of underlying mathematical structures and functions, contributing significantly to the foundations of order theory, uncertainty modelling and aggregation theory. Computational modelling at KERMIT is dedicated to developing and applying cutting-edge techniques—such as differentiable, probabilistic, and evolutionary computation—to enhance the understanding and engineering of biological systems. By integrating AI-driven simulations, the team bridges the gap between theoretical models and real-world applications.

On Friday, November 6, the members of the KERMIT research unit hosted a hackathon as a team-building exercise. The aim was to learn to solve bioscience engineering problems using modern generative AI tools, while keeping the human in the loop.

The first group was challenged to develop a diverse recipe generator. The project built on a research idea that frequently comes up in our group: How can we find diverse solutions (e.g., in terms of ingredients and cuisines) while still maintaining a high level of quality (in this case, taste)? Ultimately, the group developed a system that can generate a diverse Christmas menu while accounting for different people's allergies. In addition, the system aimed to determine an optimal preparation order for the different dishes. From a mathematical point of view, this challenge is closely related to finding efficient and sustainable production processes for chemicals and bioproducts.

The second group focused on developing an automated paper-screening pipeline, built entirely on open-source models and databases. Given a research question, the goal was first to identify the most relevant papers and then screen these papers further to determine which sections were of interest. We made this pipeline safe, reliable, and computationally efficient by running a small local model through Ollama as the judge, instead of sending every paper to a large cloud-hosted model. Similar setups could automatically annotate phage proteins at scale or support trustworthy medical diagnoses.

We concluded the day with short presentations of the results. Overall, we were impressed by what we achieved in just one day. We found that mornings of brainstorming with pen and paper, followed by AI-aided coding in the afternoon, are an effective way to build working prototypes in a day. These insights are highly valuable in modern academic teaching and research.