Metadata Agents in Practice: Collaborative Design, Deployment, and Evaluation in a Production Repository System
Brendan Quinn
Lead AI Engineer
Northwestern University
Northwestern University Libraries will present work on developing metadata augmentation agents for its digital asset repository, Meadow, supported by an Institute of Museum and Library Services (IMLS) National Leadership grant. The presentation will explore how the Library is integrating Anthropic’s Claude Agent SDK with Amazon Web Services, including Amazon Bedrock and OpenSearch, to create a cloud-based system that enhances metadata creation and enrichment. The briefing includes lessons from the design process, addressing technical challenges and approaches to building trust in artificial intelligence tools for metadata work, as well as experiences on integrating language models with repository workflows, agent design strategies, staff engagement, and plans for future expansion across collections.
https://github.com/nulib/meadow
https://www.imls.gov/grants/awarded/lg-256703-ols-24
Striking the Right Balance: Evidence-Based Strategies for Implementing Artificial Intelligence in Library Metadata Workflows
Kate James
Metadata Engagement Program Coordinator
OCLC
Jill Reilly
Acting Chief Innovation Officer
National Archives and Records Administration
The OCLC Research Library Partnership’s Managing AI in Metadata Workflows Working Group brought together metadata managers from academic libraries, archives, and museums to examine real-world AI implementation across cataloging, special collections, and institutional repository workflows. Through collaborative investigation of specific use cases—from addressing cataloging backlogs to supporting accessibility requirements—the group identified both significant opportunities and critical implementation challenges. The session will present evidence-based findings on three critical topics: where AI shows genuine promise for addressing capacity constraints, what quality frameworks are essential for responsible adoption, and how to maintain professional expertise while leveraging AI capabilities. Participants emphasized the value of specialized over general-purpose AI tools, the need for methods to ensure metadata quality and transparency, and the importance of implementing models that enhance rather than replace human judgment. The discussion will focus on practical strategies for responsible AI integration, emerging best practices, and frameworks for evaluating AI tools in metadata contexts. The session will also include details from NARA’s Citizen Archivist program, which supports volunteers in tagging and transcription of digitized resources alongside AI generated metadata, as an example of where humans and machines are needed and mutually beneficial for achieving large, complex tasks.
https://www.oclc.org/research/partnership/working-groups/managing-ai-metadata-managers-wg.html
https://hangingtogether.org/tag/aiandmetadataworkflows/
Presentation (James)
Presentation (Reilly)
