LLMs and Retrieval Augmented Generation (RAG) for libraries

Research, perspectives and valorisation of heritage

Abstract

This paper explores the opportunities offered by the adoption of Retrieval-Augmented Generation (RAG) in libraries, with a focus on the enhancement of personal archives and authors' libraries. The aim is to investigate how the integration of artificial intelligence, documentary description, and narrative practices can improve access, contextualization, and the cultural use of complex heritage materials. After a reflection on the role of generative technologies in library services, the article presents a case study developed at the “Arturo Graf” Library of the University of Turin, centered on the Emanuele Artom collection. The project promotes an interdisciplinary and narrative-driven approach to highlight the relational and cultural value of personal collections, fostering a participatory and meaningful engagement with documentary memory.

The authors share responsibility for developing the content of the contribution as a whole. Lorenzo Verna authored the sections “I Large Language Models: definizione, architettura e limiti” and “L’architettura Retrieval-Augmented Generation (RAG).” Roberto Testa authored the section “LLM e RAG in Biblioteca: applicazioni e prospettive,” and Angelo La Gorga authored the section “Verso la RAG: la valorizzazione dei fondi di persona come possibile campo di applicazione.” The introduction and concluding section were jointly written.