Tutorial Generation for Code Repository based on Multi-agent System
Bachelor & Master Thesis
We are launching a research project on automatic tutorial generation for codebases using multi-agent collaboration. Existing approaches, including documentation tools and large language models (LLMs), often fall short in producing structured, pedagogically effective tutorials. They typically rely on single-agent generation pipelines that lack interactivity, overlook user intent, and struggle with capturing the complex dependencies and modular structure of real-world software systems.
The motivation behind this work stems from the growing complexity of modern codebases and the increasing demand for intelligent systems that can help users—especially newcomers—understand, navigate, and learn from large-scale software projects. Automatically generating tutorials that are both informative and adaptable to different user profiles remains a highly challenging task. It requires not only deep semantic understanding of the code but also the ability to organize knowledge in a coherent and goal-oriented way. Designing a system where multiple agents take on different roles—such as explainer, questioner, and demonstrator—and interact to iteratively refine the tutorial content presents a novel and promising direction.
Our method leverages a multi-agent system architecture in which different agents collaborate to generate high-quality, customized tutorials. Each agent is designed with distinct goals and strategies, enabling richer context modeling, cross-module reasoning, and more natural instructional flow. This setup also allows us to incorporate user feedback and dynamically adapt the tutorial content based on different learning goals or user expertise levels.
Students involved in this project will have the opportunity to work with multi-agent systems and large language models in the context of code understanding and tutorial generation. Through hands-on implementation and experimentation, they will develop a deeper understanding of how to design collaborative AI agents and apply them to practical software engineering problems. Participants may also gain experience in conducting empirical evaluations and writing technical reports or research papers, depending on the progress and outcomes of the work. This project offers a chance to explore a timely research topic at the intersection of AI and software engineering in a supportive, research-oriented environment.