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

What changed This paper presents Earthquaker-AI, an educational framework that builds upon a prior robotics project by incorporating a conversational AI assistant powered by Retrieval-Augmented Generation (RAG). The system is designed to enhance earthquake preparedness and promote conscious action among primary school students. It extends the existing "Earthquaker" STEM project, which previously focused on mechanical simulation using Lego WeDo2, to include cognitive and metacognitive processing. The robotics component utilizes Lego WeDo2 automation to simulate seismic responses, allowing students to interact with sensors and actuators as tangible representations of protective actions. The AI assistant functions as a guided learning mechanism, aligning student responses with safety guidelines and providing verbal feedback based on a rubric. This feedback is intended to support self-regulated learning and foster calmness during emergency simulations.

The Earthquaker-AI framework follows a progressive learning trajectory tailored to cognitive development. For younger grades, the focus is on basic recognition of safety actions through multiple-choice questions, assessed using a two-dimensional rubric. In middle grades, students identify correct action sequences via multiple-choice questions, evaluated with a three-axis rubric. For upper grades, the system shifts to verbal production, requiring short written responses that are assessed using a four-dimensional rubric that includes clarity of expression. The dialogic module employs RAG to semantically match student queries with official guidelines, thereby generating safe and accurate responses. Experimental evaluations indicate high groundedness and accuracy, with a low rate of hallucinations.

Why it matters for builders For AI builders and educators, Earthquaker-AI demonstrates a practical application of RAG in an educational context, specifically for teaching critical safety procedures. The integration of RAG with a robotics simulation provides a blueprint for creating interactive learning experiences that are both informative and engaging. The rubric-based assessment system offers a structured method for evaluating student understanding and self-regulation, which can be adapted for other AI-driven educational tools. Furthermore, the framework's emphasis on cognitive and metacognitive processing highlights opportunities for developing AI systems that support deeper learning and skill acquisition.

Practical impact The Earthquaker-AI system combines hands-on engagement with robotics, information processing through AI, and reflective practice via rubric-based feedback. This holistic approach aims to promote technological literacy, self-regulation, and the responsible use of digital systems among young learners. By equipping students with early crisis-management skills, the framework contributes to building a more prepared and resilient generation. The RAG component ensures that the AI assistant provides accurate and contextually relevant information, crucial for safety education.

Caveats and source limits The provided source is a research paper abstract, which offers a high-level overview of the Earthquaker-AI framework. Specific technical details regarding the RAG implementation, the exact nature of the Lego WeDo2 integration, the algorithms used for rubric-based assessment, and the quantitative results of the experimental evaluation are not fully elaborated. The paper mentions experimental evaluation showing high groundedness and accuracy with a low hallucination rate, but the specifics of this evaluation are not detailed. The source also refers to a previously implemented educational robotics project, but details of that project are not provided here. The target audience is primary school students, and the framework's adaptability to other age groups or subject matters is not discussed.