RoboTuB: Retrieval-Augmented Tutoring for Adaptive Learning in STEM MOOCs

Abstract

Educational AI chatbots have shown promise in enhancing learning, yet they often lack adaptive reasoning and pedagogical scaffolding for domain-specific subjects. This paper presents RoboTuB, a Retrieval-Augmented Generation (RAG)-powered tutoring chatbot, designed to support team-based learning in STEM education. Unlike traditional question-answering (QA) systems, RoboTuB retrieves structured knowledge from domain-specific corpora, dynamically generating explanations, simulations, and code examples tailored to learners' needs. We evaluate RoboTuB in a two-phase study: (1) a usability assessment with novice learners using System Usability Scale (SUS) and NASA-TLX cognitive load metrics, and (2) a comparative analysis in a MOOC setting, where students using RoboTuB showed statistically significant learning gains over those relying on traditional Q&A forums. Our findings underscore the value of retrieval-augmented tutoring in reducing cognitive load, improving engagement, and enhancing STEM learning outcomes. We discuss implications for NLP-driven educational chatbots and propose future directions for multimodal AI tutors.