Wei Xuan
Associate University Librarian
University of Manitoba
Mullai Manickavalli
AI Developer
University of Manitoba
As academic libraries navigate the generative artificial intelligence (AI) landscape, deploying production-ready conversational agents requires balancing emerging technical architectures with cross-departmental staff governance and rigorous quality evaluation. This briefing details the multi-year trajectory of developing and evolving an institutional AI chatbot at the University of Manitoba Libraries. From the outset, the Libraries prioritized staff engagement by forming a dedicated chatbot review group composed of library staff to establish evaluative criteria, assess service impacts, and build institutional trust. The presenters will explore the initial design and pilot of the first iteration, built on a traditional Retrieval-Augmented Generation (RAG) architecture grounded in internal library documentation, and how the staff review group evaluated answer accuracy and relevance to determine operational readiness. They will then trace the transition to Chatbot 2.0, which migrates the system from static text-retrieval RAG pipelines to the Model Context Protocol (MCP). By leveraging MCP host-server standards, the chatbot moves beyond answering static frequently asked questions to dynamically interfacing with live, external library tools, scheduling services, and real-time application programming interfaces. Attendees will gain practical, evidence-based insights into orchestrating staff-driven AI evaluation workflows and technical lessons learned from adopting MCP in library infrastructure.
