Collaborative research introducing RAP and RouteabilityBench to assess the potential benefits of LLM routing has been accepted at NeurIPS 2026 in the Evaluations and Datasets Track.

Following its NeurIPS 2026 acceptance, the AutoLearn-SI tutorial on robust, evidence-based evaluation of foundation models has also been accepted at IE 2027.

An international team of researchers presents how artificial intelligence is transforming food, nutrition, and health in a new chapter published online by Elsevier.

AutoLLMSelect results on robust and evidence-based foundation-model evaluation will provide a foundation for an accepted NeurIPS 2026 tutorial.

Research results achieved within AutoLLMSelect were spotlighted in an invited talk at AutoAI-FM 2026, highlighting advances in robust evaluation, efficient benchmarking, and LLM selection and routing.

Dr. Ana Gjorgjevikj has started a two-month MSCA research secondment at the University of Freiburg to advance automated LLM selection and benchmarking of foundation models for tabular data.

At COSEAL 2026, Ana Gjorgjevikj presented new research from the AutoLLMSelect initiative on the robustness of multilingual embedding model rankings across languages, tasks, and evaluation methodologies. The study demonstrates that conclusions about top-performing multilingual models can strongly depend on dataset composition and ranking design choices, highlighting the need for more transparent and robustness-aware benchmarking practices.

Researchers from the Jožef Stefan Institute presented the FoodBench-QA shared task at the CL4Health Workshop collocated with LREC-COLING 2026, showcasing advances in trustworthy and semantically grounded AI for food and nutrition question answering.

FoodBench-QA, accepted at LREC 2026, introduces a benchmark for grounded food and nutrition question answering, highlighting strong performance across tasks while emphasizing the need for trustworthy, explainable, and regulation-aware AI systems.

Our newly accepted paper introduces FoodyLLM, a domain-specialized large language model for nutrient estimation, traffic-light multi-label classification, and ontology-based food entity linking, significantly outperforming general-purpose LLMs across all tasks. The work highlights the importance of trustworthy benchmarking (AutoLearn-SI) and sustainability-aware model selection (AutoLLMSelect), demonstrating that smaller domain-adapted models can achieve higher accuracy with lower computational cost.

BFNDMA 2025 at IEEE Big Data showcased cutting-edge AI research in food and nutrition science, featuring advances in nutrient estimation, recipe analysis, food knowledge retrieval, and domain-specific language technologies.

FoodBench-QA 2026 is a shared task on grounded food and nutrition question answering, challenging systems to perform nutrient estimation, FSA traffic-light prediction, and food entity recognition/linking using structured dietary resources. Hosted at CL4Health @ LREC 2026, the competition provides realistic food-related queries and invites participants to submit system papers following the evaluation phase.

Ana Gjorgjevikj presents FoodSEM, a state-of-the-art open-source LLM for food named-entity linking, at Discovery Science 2025.

The paper studies the challenge of associating sustainable development indicators used at different levels of governance (e.g., local, regional) to the UN Sustainable Development Goals (SDGs) and their targets, in order to better align the different efforts towards sustainable development with the SDGs.

Our article, accepted by Knowledge-Based Systems, presents xLLMBench, a transparent framework for ranking LLMs using multi-criteria decision-making. Evaluated on HuggingFace and HELM Classic leaderboards, it enables application-specific benchmarking, revealing how rankings shift under different metrics for transparent and reproducible insights.

📢 BFNDMA 2025 invites research on Big Data, AI, and NLP solutions for food, nutrition, and sustainability—submit your work and join us in Macau this December!

An open-source large language model for food named-entity linking, FoodSEM advances multi-label classification and sustainable AI, showcasing the synergy between the AutoLLMSelect and AutoLearn-SI projects.