Empowering Online STEM Labs with Generative AI: Real-Time Anomaly Detection and Adaptive Support Using LLMs

OELAssist is a prototype AI-powered support system developed to enhance remote STEM education by providing real-time, personalised assistance through generative AI and large language models (LLMs). Designed for use within the Open Engineering Lab (OEL) at The Open University, the system monitors student interaction data during virtual lab sessions to identify signs of struggle or confusion. Using an anomaly detection model such as Isolation Forests, alongside classification models like Random Forests, the system identifies and classifies atypical behaviours and passes this information to a Retrieval-Augmented Generation (RAG) pipeline, which prompts an LLM to generate context-aware feedback for learners. Currently in its pilot and testing phase, OELAssist is being trialled using historical and live experimental data from student engagement with pressure vessel activities. In the short term, the system aims to demonstrate the feasibility and potential impact of automated, AI-driven feedback to support learners in real-time, particularly in remote lab settings where instructor presence is limited. In the longer term, if proven effective, the system could be scaled and integrated into remote STEM modules across disciplines, offering adaptive, data-informed support that helps reduce student frustration, close awarding gaps, and improve retention. This work highlights how generative AI, when combined with anomaly detection, can inform innovative, student-centred approaches to online lab education.