Implementing Retrieval-Augmented Generation for Academic Libraries

This article details the technical development of a Retrieval-Augmented Generation (RAG) system designed to enhance discovery within an academic library's institutional repository. Conducted during a six-month research leave in 2025, this project explores the practical application of emerging cloud-based AI tools in a library context. We developed a prototype that integrates the University of Manitoba’s MSpace repository with Microsoft Azure AI services. The system utilizes an OAI-PMH harvester to retrieve metadata, generates semantic vector embeddings via the text-embedding-ada-002 model, and indexes these vectors in Azure AI Search. A custom front-end application facilitates both traditional keyword search and generative, context-aware chat interactions. This paper documents the development environment, script logic, and specific technical challenges overcome—such as OAI-PMH pagination errors and API versioning conflicts—providing a reproducible roadmap for libraries seeking to explore semantic search technologies.