Copilot Tutor: Automated Software Engineering Practice Augmented with LLMs

Abstract

In recent years, large language models (LLMs) have become more ubiquitous in the workplace. In software engineering, they are often realized as “copilots" which produce code given a prompt or existing code. Programmers using these tools to increase their coding productivity need to be proficient in inspecting and in understanding these copilots’ outputs. As engineers incorporate these tools to accelerate their workflows, they have a parallel opportunity to accelerate learning new programming languages. This thesis presents a tutor interface where students with some programming experience in an origin language can learn a target language while practicing how to critically read and fix a copilot’s output to write correct, safe programs. This work also introduces the automatic generation of exercises teaching syntax and semantics on which a programmer experienced in the origin language but not the target language should focus.