Earthquaker-AI: A Retrieval-Augmented Generation Framework with Rubric-Based Assessment for Primary School Earthquake Education

Earthquake preparedness education in primary schools often suffers from being overly theoretical and lacking individualized, formative feedback. While educational robotics (such as the original Earthquaker project using Lego WeDo2) successfully engages students in physical simulations of seismic responses, these mechanical activities often fail to address the cognitive and metacognitive aspects of safety. Earthquaker-AI evolves this framework by adding a conversational AI layer that guides students through earthquake safety scenarios, helping them move from physical execution to structured reasoning and verbal articulation.

The system utilizes a Retrieval-Augmented Generation (RAG) architecture to ensure that all AI-generated responses are strictly grounded in official earthquake safety guidelines. By using semantic embeddings to retrieve relevant excerpts before generating an answer, the system minimizes the risk of hallucinations. The framework operates in two modes: a 'Question Mode' for open inquiry and a 'Quiz Adventure Mode' that presents grade-appropriate scenarios. These scenarios are evaluated using explicit, age-specific rubrics that assess everything from basic action recognition in early grades to complex justification and clarity of expression in upper grades.

Earthquaker-AI is designed to support a progressive learning trajectory aligned with cognitive development. By providing non-punitive, rubric-based feedback, the system acts as a pedagogical mediator rather than an authoritative evaluator. This approach encourages students to reflect on their decision-making processes, fostering self-regulation and calmness under pressure. The integration of robotics, AI, and formative assessment aims to build technological literacy while ensuring students internalize critical crisis-management skills.