LibraryAI — Explorative Development and Testing of a Virtual Library Research Assistant Application

• Applicants: Professor Dr.-Ing. Bela Gipp; Professor Dr. Julian Kunkel
• Term: since 2025
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 559325635

Over the last decades, libraries have undergone major technological transformations to enhance the accessibility and management of their collections and services. Large language models (LLMs) like OpenAI's ChatGPT represent the next technological leap that opens vast opportunities for libraries. Virtual assistants based on LLMs have drastically pushed the boundaries of natural language processing and understanding. However, today’s LLMs exhibit shortcomings critical for their application in the library context either due to inherent limitations or a lack of optimisation for relevant domains and languages. To fill this gap, this project will develop, deploy, and evaluate LibraryAI—a specialized virtual library assistant leveraging LLM technology. The project will optimise access to and utilisation of the library's scientific resources and services aiming for an enhanced user experience for researchers across all disciplines. The virtual assistant will offer an interactive dialogue interface through which users can engage with the library's extensive collection of scientific resources. By chatting with the assistant, users will be able to retrieve information, ask questions, generate summaries, and translate content. Additionally, the virtual assistant will promote user engagement through conversations and thought-provoking discussions on collection material, offering a vibrant and interactive learning experience and invaluable support for researchers in their literature-centric endeavours. To achieve these goals, the project will employ transfer learning to adapt open-source LLMs to the library’s domains and languages. We will address LLM limitations—such as outdated information or hallucinations—by integrating the virtual assistant with information retrieval systems and existing library APIs. The project is a collaborative effort of the SUB Göttingen, the Scientific Information Analytics Group at the Institute of Computer Science of the Georg-August-University Göttingen, and the GWDG computing centre. Our multidisciplinary team brings together expertise in library sciences, computer science, and data management. Moreover, the project can leverage significant data and high-performance computing resources. This blend of domain-specific knowledge, technical skill, and data infrastructure will empower us to devise, implement, and deploy a capable, robust, and efficient language model along with a user-friendly interface. LibraryAI will be agnostic to the underlying library systems and collections, feature a modular architecture and be made openly accessible to ensure its adaptability for other libraries.

• DFG Programme: Research data and software (Scientific Library Services and Information Systems)