#1352: STEPS: An AI Tutor for Scaffolded Problem Solving in Introductory STEM Courses

Introductory STEM courses often reveal a disconnect between taught content and students’ ability to solve complex, multistep problems. We present STEPS—an LLM-based tutor drawing on principles of deliberate practice—that scaffolds planning, step-by-step reasoning, and verification without giving away answers. Through iterative prompt engineering aligned with a validated problem-solving framework, STEPS was embedded into weekly assignments, quizzes, reflections, and lectures in large-enrollment chemistry and physics courses. We examined plan quality, solution correctness, student engagement, perceived helpfulness and confidence, and performance on targeted concepts. Results show high confidence, strong homework performance on STEPS-assisted problems, and increasing perceived helpfulness, with students shifting from answer-seeking to process-oriented inquiry and demonstrating growing AI literacy. This session will provide a brief overview of the framework, hands-on use of STEPS, and a guided workshop to craft authentically complex problems and translate them into STEPS-style prompts across STEM disciplines. We conclude with actionable guidance for pedagogically sound, learning-aligned AI integration in STEM courses.