FAIR Data Assessment Using LLMs: The Fair-Way

As part of modern research practices, the FAIR data principles have become essential for data discoverability, usability, and sharing. Existing implementations for automatically assessing FAIR adherence (FAIRness) often suffer from limited usability, inconsistent accuracy, and difficult-to-interpret results, as they require explicit rules to cover for specific FAIR assessment frameworks, which are not easy to generalize. This paper introduces Fair-Way, an open source tool that leverages Large Language Models (LLMs) to automate FAIRness assessment. Fair-Way applies a divide-and-conquer approach to decompose the assessment process into fine-grained tasks, as well as to split the metadata into manageable chunks. Evaluation demonstrates that Fair-Way achieves performance comparable to existing tools, while outperforming them in several key metrics. Moreover, Fair-Way generalizes across FAIR assessment indicators without requiring explicitly programmed logic and supports both structured and unstructured metadata in diverse formats. Finally, it enables user-defined, domain-specific tests, which are typically not supported by other systems. Overall, Fair-Way represents a scalable and flexible solution to accelerate FAIR data practices across research domains.