GPT Teacher: desenvolvimento de um agente de LLM para programação assistida em ambiente VSCode

Abstract

The computer programming teaching and learning process presents significant challenges that often result in comprehension difficulties and student demotivation. This paper presents the development and evaluation of GPT Teacher, an LLM-based agent for assisted programming integrated into the Visual Studio Code (VSCode) environment. The primary objective was to design and implement an LLM-based tool focused on programming education, capable of supporting the acquisition of programming competencies more effectively than generic coding assistants. The methodology involved creating a functional prototype based on a dual-agent architecture: a Diagnostic Agent, responsible for technical code analysis, and a Guidance Agent, tasked with translating this analysis into a constructive and educational dialogue for the student. The results of the functional validation demonstrate that the proposed approach is robust and promising, confirming the hypothesis that LLM agents, when structured within a specialized system, can serve as powerful allies in the programming teaching-learning process by reconciling technical rigor with pedagogical effectiveness.