Enhancing self-regulated learning with large language models

A pilot study on the feasibility of local deployment

Abstract

Self-regulated learning (SRL) is essential for academic success, yet many learners struggle to plan, monitor, and reflect their learning processes without support. Large Language Models (LLMs) offer opportunities for real-time, personalised learning guidance, but cloud-based deployments raise privacy and trust concerns. This pilot study investigates the feasibility of delivering SRL support through a locally deployed, privacy-preserving chatbot. Using a design-based research approach, we co-designed a chatbot platform with sixty-one university students and conducted a two-week field study with seven participants using both local (offline) and cloud-based (online) modes. Mixed-method findings indicate that the chatbot successfully prompted higher-order SRL activities such as goal setting and reflective monitoring in authentic study sessions. Participants reported greater trust when using the fully local LLMs due to data remaining on-device. However, the local LLMs demonstrated much slower response times and occasional inaccuracies, highlighting privacy-performance trade-offs. This research demonstrates the potential of locally deployed, privacy-preserving, human-centred AI to support SRL and offers empirical insights into the benefits and limitations of deploying LLMs on small-scale local devices in educational contexts.