Virginia tech university libraries will plan an investigation of the auditability of large language models for relevance screening in evidence synthesis

The Institute of Museum and Library Services awarded Virginia Polytechnic Institute & State University (Virginia Tech) $357,988 on September 1, 2026, under the National Leadership Grants program (CFDA 45.312) to plan an investigation of the auditability of large language models for relevance screening in evidence synthesis.

Virginia Tech's University Libraries will conduct iterative experiments to improve the accuracy and auditability of AI-assisted screening workflows used in evidence synthesis. The project incorporates feedback from evidence synthesis library professionals across the nation and will produce an AI-enabled evidence synthesis prototype, an instruction manual, and a feasibility report with implementation guidance. The deliverables are designed to help libraries assess how AI-assisted screening can be integrated into evidence synthesis services while preserving established methodological standards.

Performance is based in Blacksburg, Virginia, with a period of performance extending through August 31, 2028. The National Leadership Grants program supports projects addressing critical needs of the museum and library fields with the potential to generate results such as models, new tools, research findings, services, practices, and alliances that can be widely used, adapted, scaled, or replicated to advance practice in these professions and strengthen services for the American public.

This project will run iterative experiments to improve the accuracy and auditability of the artificial intelligence-assisted screening workflow. The project will incorporate the feedback from evidence synthesis library professionals across the nation and will produce feasibility findings, prototype workflows, and implementation guidance that libraries can use to assess how artificial intelligence-assisted screening can be incorporated into evidence synthesis services while preserving established methodological standards. Deliverables include an artificial intelligence-enabled evidence synthesis prototype; an instruction manual; and a report.