Course-Specific AI Chatbots for Universities

A course-specific chatbot can combine a language model with retrieval from selected course materials. A student asks a question; the system searches approved content such as the syllabus, lecture notes, readings, and assignment instructions, then drafts an answer grounded in relevant passages. The Jill Watson project at Georgia Tech is an example of a course assistant built around instructor-approved courseware and retrieval-supported responses. Course grounding can reduce unsupported answers, but it does not guarantee correctness. A bot may retrieve an outdated policy, miss an exception, or misinterpret a technical passage. Keep source documents versioned and remove or replace old materials when course policies change. Show the source used, label the answer as a suggestion, and provide a direct path to the instructor or teaching assistant. Define what the assistant may do. It can help locate readings, explain terms from class materials, and answer routine logistics. It should not assign final grades, make accommodation decisions, resolve academic-integrity disputes, or provide mental-health or financial advice. When a student's circumstances matter, escalate rather than infer from chat history. The course team remains responsible for policy interpretation and educational judgment. Evaluate the bot before launch with representative questions from prior courses or instructor-created test cases. Include ambiguous prompts, outdated notes, unsupported topics, and attempts to obtain another student's information. Check whether answers are faithful to retrieved passages, citations point to the right materials, and the bot appropriately says when it does not know. Review failures after the course begins. Student questions can contain identifiable education records or sensitive information. Use institution-approved services, minimize data collection, restrict access, and review retention, model-training, and vendor-sharing terms. Follow institutional policy and applicable privacy requirements. A course chatbot should supplement teaching presence, not replace human support.