#1317: SEED: A Teachable-Agent System that Supports Elementary Students’ Self-Explanation by Teaching an AI Chatbot

Self-explanation is a powerful strategy for conceptual understanding, yet elementary students struggle to produce coherent explanations without individualized scaffolding. LLM-based teachable agents offer a promising alternative—positioning the learner as teacher and the AI as student, leveraging the protégé effect to deepen understanding—yet most existing work targets undergraduate learners, leaving elementary students largely unaddressed. This demo introduces SEED (Self-Explanation Eliciting Dialogue), a prototype teachable-agent system that positions the child as the teacher and the AI as a novice learner. SEED features a role-reversal prompt engine, three-stage scaffolding flow, growth visualization, and auto-generated learning notebooks—all designed to sustain the protégé effect. This work demonstrates that LLM-based teachable agents, when designed around developmental principles, can meaningfully support self-explanation among elementary learners.