Posters

This study examines how the cataloging performance of four AI models, ChatGPT, Copilot, DeepSeek, and Gemini, evolved over eight months when tasked with extracting bibliographic information from scanned images across seven items of varying publication types and subject domains. Using four prompt variations and a consistent methodology established in an earlier round of testing, the second round revealed meaningful overall improvement in the accuracy and completeness of cataloging records, with models more consistently acknowledging missing information, providing inline justification for decisions, and exhibiting behaviors aligned with Explainable AI (XAI) and Retrieval-Augmented Generation (RAG) principles. Persistent challenges remained in controlled subject headings and URI accuracy, and a new concern emerged around balancing prompt over- and under-specification. These findings support a human-in-the-loop approach to AI-assisted cataloging and highlight the value of continued longitudinal monitoring.

This poster introduces SAMATA (समता), a layered metadata application profile designed to represent knowledge resources and their transmission across textual, archival, print, oral, and material forms. While standards such as Dublin Core, CIDOC-CRM, and PREMIS provide robust infrastructures for describing information objects, they remain limited in representing collections embedded within complex knowledge systems and community-governed contexts. Drawing on field-based metadata work across manuscript, archival, print, and museum collections, this study identifies three recurring structural gaps: (1) limited representation of non-Gregorian temporal systems; (2) absence of structured modelling for intellectual transmission relationships; and (3) lack of machine-readable governance metadata addressing community consent and ethical use, including AI training conditions. SAMATA addresses these gaps through a sixteen-layer architecture that extends existing standards while maintaining interoperability. The framework is currently under pilot implementation at Chandernagore College under a formal institutional MoU.