# RAG on Institutional Knowledge: Catalogs, Policies, Handbooks > Higher Education · AI Course · HE-7 > Source: https://ibl.ai/solutions/higher-education/course/rag-on-institutional-knowledge > Last updated: 2026-08-25 **Retrieval-augmented generation over the documents a campus runs on — chunking a catalog, versioning policy, and stopping the agent citing a 2019 handbook.** ## The Short Answer **Campus retrieval is hard because rules are effective-dated and documents contradict each other. ibl.ai builds RAG that answers by catalog year, cites the governing document, and surfaces contradictions instead of averaging them — running on infrastructure where you own all the code and the data, so policy documents and student queries stay inside the institution.** On ibl.ai you own all the code and the data, run it model-agnostic across any LLM, and pay with no per-seat pricing — so you can deploy anywhere, from your own cloud to a fully air-gapped network. [Request Access](https://ibl.ai/contact) · [Explore Higher Education](https://ibl.ai/solutions/higher-education) ## Course facts - **Level:** Advanced - **Duration:** 7 hours across 8 modules - **Format:** Technical workshop with a build lab - **Modules:** 8 - **Catalog code:** HE-7 - **Frameworks covered:** FERPA, NIST AI RMF ## What is this course about? A campus is a genuinely hard retrieval target: rules are effective-dated, documents contradict each other, and the authoritative answer depends on when the student enrolled. This course builds RAG that handles all three, with citation and provenance so a student can verify the answer and a registrar can defend it. ## Who is this course for? - Academic technology engineers and developers - Data and integration architects - Registrars partnering on catalog structure - IT leaders evaluating build versus buy ### What do I need before starting? - Comfort with APIs and structured data - Basic familiarity with embeddings and vector search ## What will I be able to do afterwards? - Chunk structured institutional documents without destroying their meaning - Implement effective-dated retrieval that answers by the student's catalog year - Detect and surface contradiction between authoritative documents - Attach citation and provenance so any answer can be verified - Build a registrar-authored gold set and evaluate retrieval against it ## What does each module cover? ### Module 1 — Why is a campus a hard retrieval target? The three properties — effective dating, exceptions, contradiction — that break naive RAG on institutional content. _(45 min)_ **Objectives** - Characterize the properties that distinguish campus content - Predict where a naive implementation will fail - Set evaluation criteria before building **Topics:** Effective dating · Exception handling · Document contradiction · Failure prediction **Activity:** Take ten real policy questions and predict which a naive RAG system would get wrong. ### Module 2 — How do you chunk a course catalog? Chunking structured documents — requirement trees, tables, and cross-references — without severing meaning. _(60 min)_ **Objectives** - Choose chunking strategies by document structure - Preserve table and requirement-tree integrity - Handle cross-references between chunks **Topics:** Structure-aware chunking · Table preservation · Requirement trees · Cross-reference resolution **Activity:** Chunk a real catalog section three ways and compare retrieval quality on the same questions. ### Module 3 — How do you answer by catalog year? Effective-dated retrieval — the campus-specific capability no general RAG tutorial covers. _(60 min)_ **Objectives** - Model effective dates as retrievable metadata - Filter retrieval by the student's applicable edition - Handle questions that span multiple editions **Topics:** Temporal metadata · Edition filtering · Multi-edition questions · Default behavior when unknown **Activity:** Implement edition filtering and verify against a question set with known per-year answers. ### Module 4 — What happens when the catalog and the handbook disagree? Contradiction detection and authority ranking, rather than silently blending conflicting sources. _(55 min)_ **Objectives** - Establish a document authority hierarchy - Detect contradiction between retrieved chunks - Surface the conflict rather than resolving it silently **Topics:** Authority hierarchy · Contradiction detection · Conflict surfacing · Escalation to a human **Activity:** Seed a deliberate contradiction and verify the system surfaces rather than averages it. ### Module 5 — How do you attach citation and provenance? Making every answer verifiable — the property that determines whether staff will trust the system. _(50 min)_ **Objectives** - Attach source document, section, and effective date to every answer - Render citations a student can act on - Detect when an answer is not adequately grounded **Topics:** Provenance metadata · Citation rendering · Groundedness detection · Abstention behavior **Activity:** Add provenance to your pipeline and test abstention on questions with no grounded answer. ### Module 6 — How do you build a gold set registrars trust? Evaluation data authored by the people who know the correct answers. _(55 min)_ **Objectives** - Recruit domain authors and structure their contribution - Write gold questions with defensible correct answers - Version the gold set as policy changes **Topics:** Domain-authored evaluation · Answer defensibility · Gold set versioning · Inter-annotator agreement **Activity:** Run a gold-set authoring session with a registrar and produce 30 verified items. ### Module 7 — What breaks when the catalog is republished? Refresh pipelines and the regression testing that catches silent degradation. _(50 min)_ **Objectives** - Design an ingestion pipeline for document republication - Detect regression after a content refresh - Handle partial and rolling updates **Topics:** Ingestion pipelines · Regression detection · Partial updates · Rollback **Activity:** Republish a modified catalog and run the regression suite to catch what degraded. ### Module 8 — Building a policy agent that cites its source The hands-on module: a complete policy agent with effective dating, contradiction surfacing, and provenance. _(75 min)_ **Objectives** - Assemble the full pipeline end to end - Pass the gold set at an agreed threshold - Document the architecture for handover **Topics:** Pipeline assembly · Threshold evaluation · Architecture documentation · Operational handover **Activity:** Deploy the complete agent and pass the registrar-authored gold set. ## What is the capstone project? **Institutional policy agent with a registrar-verified gold set.** Build a complete retrieval system over your institution's catalog and handbook, with effective-dated retrieval, contradiction surfacing, provenance, and a registrar-authored evaluation set it must pass. _Deliverable:_ A deployed agent, its architecture documentation, and a versioned gold set with passing scores. ## How are learners assessed? - Gold set performance against a pre-agreed threshold - Contradiction handling verified with seeded conflicts - Architecture documentation reviewed for handover sufficiency ## What ships with the course? - **Facilitator guide.** Session-by-session running order, discussion prompts, and the questions that reliably derail a room. - **Learner workbook.** Exercises, checklists, and the templates each module's activity produces. - **Hands-on lab environment.** A sandboxed ibl.ai deployment so exercises run against real agents, not screenshots. - **Assessment bank.** Scenario questions and rubric criteria mapped to each stated learning outcome. - **Source bibliography.** Every primary regulation and standard cited on this page, linked and dated. ## Which AI agents does this course use? - [Administrative Agent](https://ibl.ai/solutions/higher-education/agent/administrative-agent) - [Student Services Agent](https://ibl.ai/solutions/higher-education/agent/student-services-agent) - [Academic Advisor Agent](https://ibl.ai/solutions/higher-education/agent/academic-advisor-agent) - [IT Help Desk Agent](https://ibl.ai/solutions/higher-education/agent/it-help-desk-agent) ## Where does the course material come from? Every module is grounded in primary sources — the regulation, standard, or research itself, not a summary of it. Each was resolved at authoring time. - [Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks](https://arxiv.org/abs/2005.11401) — Lewis et al., arXiv. The foundational RAG paper, used for the architectural grounding in Module 1. - [Model Context Protocol](https://modelcontextprotocol.io/) — Anthropic. Integration layer for connecting the agent to SIS and catalog systems. - [LlamaIndex](https://www.llamaindex.ai/) — LlamaIndex. Reference implementation patterns for structure-aware chunking. - [AI Risk Management Framework](https://www.nist.gov/itl/ai-risk-management-framework) — NIST. Frames the groundedness and abstention requirements in Module 5. ## Delivery notes Binding guidance for anyone preparing and delivering this course: - This is the most technical course in the higher-ed catalog and should be scoped as such. Do not dilute it for a mixed audience — run HE-1 or HE-2 for non-technical stakeholders instead. - Module 3's effective-dated retrieval is the genuinely novel content. Every general RAG course covers chunking; almost none covers temporal authority. Give it the most build time and the best examples. - The gold set session in Module 6 needs a real registrar in the room. Engineers consistently write gold questions that are too easy and miss the exceptions that matter. - Use a real catalog with at least three years of drift. A synthetic one will not contain the messy substitutions and grandfathered rules that make this hard. - Ship the regression suite from Module 7 as a reusable artifact — it is the piece institutions are least likely to build for themselves. ## Why run AI training on a platform you own? - **You own the course, not a licence to it.** Course content, learner data, and the platform run inside your perimeter — you own all the code and the data. - **Model-agnostic delivery.** Run the course's AI components on any LLM — Claude, GPT, Llama, Gemini, Command — and switch anytime. - **No per-seat training licences.** Usage-based or self-hosted, so cost tracks actual use rather than headcount. - **Deploy anywhere.** Cloud, private VPC, on-premise, or fully air-gapped — including for cohorts that cannot use public AI tools. ## Frequently asked questions ### What does the RAG on Institutional Knowledge: Catalogs, Policies, Handbooks course cover? A campus is a genuinely hard retrieval target: rules are effective-dated, documents contradict each other, and the authoritative answer depends on when the student enrolled. This course builds RAG that handles all three, with citation and provenance so a student can verify the answer and a registrar can defend it. It runs 7 hours across 8 modules across 8 modules, at advanced level, and closes with a capstone: Institutional policy agent with a registrar-verified gold set. ### Who should take RAG on Institutional Knowledge: Catalogs, Policies, Handbooks? It is written for Academic technology engineers and developers, Data and integration architects, Registrars partnering on catalog structure, IT leaders evaluating build versus buy. Prerequisites: Comfort with APIs and structured data; Basic familiarity with embeddings and vector search. ### Can we run this course on our own infrastructure? Yes. ibl.ai is model-agnostic and deploy-anywhere — cloud, private VPC, on-premise, or fully air-gapped — and you own all the code and the data. Cohort data, submissions, and any material learners upload stay inside your perimeter, which matters for higher education teams that cannot send work to a public AI tool. ### How do we get access to RAG on Institutional Knowledge: Catalogs, Policies, Handbooks? Request access and we will set it up for your cohort — hosted by ibl.ai, or running against your own deployment. Tell us the group size and timing you need, and whether it should run inside your own perimeter. ### How much does AI training for higher education cost on ibl.ai? There is no per-seat pricing — you pay for usage or self-host and pay only for the infrastructure, so a 5,000-person rollout does not cost 5,000 licences. 1.6M+ users across 400+ organizations run the platform this way, including NVIDIA, MIT, and Syracuse University. ## More Higher Education courses - [FERPA-Compliant AI: Deploying Agents on Student Data](https://ibl.ai/solutions/higher-education/course/ferpa-compliant-ai): Run AI agents against your SIS and LMS without a vendor ever seeing a student record — the school official exception, vendor DPAs, and the architecture FERPA implies. - [AI Academic Advising at Scale: Design and Guardrails](https://ibl.ai/solutions/higher-education/course/ai-academic-advising-at-scale): Build an advising agent that handles degree audits and registration at 20,000-student scale without ever giving a student wrong graduation advice. - [Enrollment and Yield AI: Agents Across the Funnel](https://ibl.ai/solutions/higher-education/course/enrollment-and-yield-ai): Deploy AI across inquiry, application, admit, and melt — where agents lift yield, where they damage trust, and how to keep the funnel on infrastructure you own. - [AI Tutoring That Improves Outcomes, Not Just Engagement](https://ibl.ai/solutions/higher-education/course/ai-tutoring-that-improves-outcomes): Design a tutoring agent that produces measurable learning gains — Socratic scaffolding, answer-withholding, misconception detection, and honest outcome measurement. - [Assessment Redesign for the AI Era](https://ibl.ai/solutions/higher-education/course/assessment-redesign-for-the-ai-era): Detection does not work. Rebuild assessment around what AI cannot fake — process, oral defense, local context, and in-class artifacts — with department-ready rubrics. - [Writing a Campus AI Policy That Survives Accreditation](https://ibl.ai/solutions/higher-education/course/campus-ai-policy-that-survives-accreditation): Draft institutional AI policy a regional accreditor, a general counsel, and a faculty senate will each accept — with the governance to keep it current.