The Problem of Institutional Amnesia
Universities generate vast quantities of knowledge — research papers, governance decisions, curriculum records, financial analyses, strategic plans — yet this knowledge is fragmented across departments, file systems, and individual memories. When a senior faculty member retires, decades of institutional context leave with them. When a governance question arises, the relevant precedent may exist in Senate minutes from fifteen years ago that no current member has read. This institutional amnesia is not merely inefficient; it is a governance risk.
RAG as Institutional Infrastructure
Retrieval-augmented generation addresses institutional amnesia by creating a queryable knowledge layer over the institution's entire documentary corpus. Documents are chunked, embedded, and stored in a vector database with rich metadata — author, date, classification, provenance chain, and content hash. When an agent receives a query, it retrieves the most relevant chunks, synthesises them into a coherent response, and provides citations back to canonical sources. The result is not a chatbot; it is a governed knowledge interface.
The Agent Architecture
At Fitzherbert University, we are designing a multi-agent system where specialised agents serve distinct institutional functions: a Governance Agent that retrieves Charter precedent and Senate decisions; a Research Agent that indexes publications and grant outcomes; an Endowment Agent that answers stewardship queries; and a Student Services Agent that surfaces policy and procedural information. Each agent operates within a defined scope, uses only authorised knowledge sources, and cannot act without human approval for consequential outputs.
Knowledge Versioning and Provenance
Institutional knowledge is not static. Policies change, research findings are updated, governance decisions are amended. A robust RAG system must therefore maintain version histories for all documents, track provenance chains that record how each piece of knowledge was created, modified, and approved, and ensure that agents always cite the current canonical version of any document. Content hashing provides tamper-evidence: any modification to a source document triggers re-indexing and provenance logging.
The Compounding Effect
The true power of agentic RAG is compounding. Every governance decision indexed, every research paper embedded, every policy update logged increases the system's institutional intelligence. Over time, the knowledge base becomes a living institutional memory that is more comprehensive, more accessible, and more reliable than any human memory or filing system. This is not artificial intelligence replacing human judgment — it is institutional infrastructure amplifying it.