Standard cloud-based LLMs often present two major hurdles for faculty: "hallucinations" that lead students away from course facts and a "give-away-the-answer" style that bypasses critical thinking. This session showcases a practical solution using Retrieval-Augmented Generation (RAG) to build course-specific AI tutors. Participants will see how instructors can serve as the "boss" of the AI by indexing their own specific materials—lecture recordings, lab manuals, and PDFs—and utilizing a pre-configured Socratic system prompt. We will share data from a Fall 2025 pilot where 91% of students reported the AI helped them identify specific weaknesses in their understanding. Attendees will leave with a blueprint for creating a disciplined AI assistant that stays grounded in their content and their teaching philosophy.
