BuddhAi: The Need to Redesign Programming as an Inquiry-based Education Through the Use of LLM Tutor

Ensuring that novice learners in undergraduate programs acquire a fundamental understanding of programming concepts is a multifaceted challenge. This challenge ranges from student knowledge diversity and the availability of tutors to the composition of exercises. It has been shown that one-on-one tutoring and interactive coding sessions can be highly beneficial to student learning, yet they are costly, do not easily scale, and require precise timing. The widespread use of large language models (LLMs) further complicates the issue, as programming exercises and hand-ins can be quickly solved using any public LLM, making it harder to continuously measure students' theoretical understanding. This early research details the motivation for developing a tutoring LLM called BuddhAi, which prioritizes Socratic dialogue over solutionism to individually assist students in their learning journeys. The LLM tutor was built using a local version of the Llama 3.1 8B model, and the user interface provides both a chat and a code window with a multi-tab workflow. The paper explores the learning problems and provides a full system overview and technical evaluations before full deployment with a student cohort. We also show the feedback from the initial stages of co-design with students.