Developing a Socratic Artificial Intelligence Tutor Using Retrieval Augmented Generation and Pedagogical Prompting
Abstract
Generative artificial intelligence (AI) applications such as ChatGPT present a distinct challenge for computer science (CS) students. While they offer individualized learning support, unregulated use risks facilitating academic shortcuts and eroding critical thinking skills. This article introduces Lion-AI, a chatbot tutor developed to augment the CS educational experience. Lion-AI integrates retrieval-augmented generation (RAG), rooted in university-specific course materials, with an engineered Socratic interaction policy. Unlike general-use tools, Lion-AI is designed to classify student requests, redirect direct-solution requests through diagnostic questioning, and provide guidance through one-question-at a-time interaction. The study describes a method for creating ethically constrained AI tutors in CS education. It presents the system’s three-tier architecture, the RAG pipeline, and the transparent interaction policy behind the tutor, positioning Lion-AI as a deployable proof of concept for combining curriculum grounding with explicit Socratic constraints. A controlled classroom study is planned as the next step.
