The Research Data Director A Persona-Based Agentic AI Blueprint and Application for FAIR Research Data, Metadata, and Schema and Crosswalk Workflows

This poster presents the Research Data Director, a persona-based architecture for applying agentic AI to the preparation, description, and publication of research data. This was presented at the Research Data Alliance p-27 Plenary as both poster and presentaion within the session 'From AI Governanece to AI Deployment Session: Trust-Centered Strategies for Agentic AI. Agentic AI infrastructures are systems in which language-model agents reason together from a situation, utilize call tools, and coordinate with one another rather than following a fixed script. They offer new capacities for the labor-intensive work of making research data Findable, Accessible, Interoperable, and Reusable (FAIR). The central design choice of this blueprint is to organize that capacity not by abstract technical function but by professional persona: each agent is modeled on a credentialed human role in a research library or institution and specified as an expert with defined competence, experience, higher education and bounded authority, and an obligation to escalate decisions beyond its scope. A coordinating Research Data Director dispatches a team of specialist persona based agents for intake, metadata generation, verification, FAIR evaluation, and provenance. The system uses two established multi-agent patterns: orchestrator-workers dispatch and an evaluator-optimizer critique loop. The environmental open ended persona approach takes advantage of LLM model intelligence to yield more coherent, auditable, and trustworthy behavior than capability-based decomposition, while going through a rigorous verification, fact checking and FAIR ethical process while keeping every irreversible act under human authority. The document specifies both the conceptual architecture and a concrete, buildable implementation — agent specification files, shared-memory design, human-checkpoint logic, and open agent protocols (MCP and A2A) — beginning with a minimal working prototype. This persona based agentic model is offered as a complementary and companion to the RDA–Microsoft Data Director and single flowchart capabilities based working-group blueprint. It is intended as an open, technology-agnostic reference that research libraries, data infrastructures, and institutions can adapt to their own environments and customize as necessary through both personas and architecture. Keywords: agentic AI; multi-agent systems; persona-based architecture; research data management; FAIR principles; research libraries; metadata; human-in-the-loop