Automating FAIR Dataset Publication Through Scientific Article Analysis

FAIRit is a tool for automatically detecting dataset usage in scientific articles, identifying missing dataset references, and generating repository-ready metadata to support FAIR data publication. The system combines rule-based identifier detection with LLM-based inference to extract both explicitly cited and implicitly described datasets. Metadata generation is constrained by a configurable schema (currently aligned with Zenodo), ensuring valid and machine-readable outputs. FAIRit supports a human-in-the-loop workflow, allowing users to review, refine, and complete metadata before publication. The tool can also automatically create and publish dataset records (e.g., on Zenodo), returning persistent identifiers and citation metadata. The repository includes the full implementation, prompts, schema definitions, and Docker setup to support reproducibility and reuse.