Hands-on RAG for Education:
From a Generic Chatbot to
a Document-Grounded System

Abstract

General-purpose chatbots can give misleading answers to advising questions when they lack an institution's current policies and catalog. This tutorial introduces computing educators to Retrieval-Augmented Generation (RAG) as a way to ground responses in institutional documents. Participants compare four RAG approaches, from no-code platforms to a custom implementation, using the same documents, test set, and scoring rubric. They then build an advising assistant from university webpages and PDFs, working through document conversion, chunking, retrieval, and answer generation. The exercises focus on citations and declining to answer when sources provide insufficient evidence. The final module considers why course recommendations require more than policy retrieval and discusses safeguards for responsible use.

By the end of the tutorial, participants will be able to

• explain how RAG grounds answers and why grounding matters for advising,
• compare four ways to build a RAG assistant in terms of effort, control, and data exposure,
• build an advising assistant from institutional HTML and PDF documents and evaluate it with a test set and rubric, and
• explain what course recommendation requires beyond document question answering and which safeguards responsible deployment needs.

No prior experience with LLMs or embeddings is required. Basic Python is helpful. Please bring a laptop with a current web browser.