Academic libraries face ongoing difficulty in aligning their collections with the needs of the communities they serve. Although the literature identified the value of using research, publications, and teaching data to inform collection development, maintaining crosswalks between these activities and library classification schemas made such work at scale impractical. This study developed a scalable workflow that maps institutional outputs to Library of Congress (LC) classification categories using a large language model to support an ongoing, evidence-based approval plan review. The workflow integrated six datasets across two domains of academic activity: research outputs and teaching activities. Where bibliographic metadata existed, they were used directly; otherwise, Anthropic’s Claude Sonnet inferred LC classifications through data-specific prompts. A Python pipeline combining the Claude and bibliographic metadata APIs produced enriched datasets at scale, followed by automated LC range validation and selective human review. The outputs feed a three-page interactive Tableau dashboard: a summary landing page, a Research Outputs Explorer, and a Teaching Activities Explorer. The collection development team has used the dashboard alongside expenditure and usage data to review approval plans, highlighting the value of unifying previously siloed data sources. The workflow is designed to refresh with new outputs and improved models, offering a sustainable foundation for ongoing collection assessment, including gap analysis and alignment with institutional priorities.
Sleeper Service: Agents as a Service. One agent. One task. A thousand of them