Building AI-Ready Scientific Data Ecosystems: From FAIR Principles to Intelligent Agent Integration
To Know
About this Class
AI-ready science is constrained less by model sophistication than by the shortage of high-quality, well-described data and by infrastructure that cannot keep pace with evolving methods. Drawing on experiences with FaceBase and the Deriva platform across craniofacial, hearing, and ophthalmology research, we argue that FAIR data largely subsumes "AI-ready" data, and that sustainable repositories follow SCALE principles: Self-service curation, domain-Agnostic platforms, Lightweight models, and Evolvable systems.
As large language model agents become the layer through which researchers interact with data, a further question arises: what makes a repository agent-ready? We propose three architectural requirements—documented service interfaces with full coverage, FAIR metadata annotated with community ontologies, and an introspectable data model with an LLM-aligned query interface—and show how an agent built on them can mediate the full research lifecycle, from discovery through citable dataset assembly to curation of results back into the repository, capturing experimental context that artifact-level reproducibility alone cannot.