Understanding an unfamiliar repository starts with basic questions: what does it do, how do its parts connect, and where should I look next? CodeMap uses visual structure and AI assistance to help developers answer those questions without losing their place in the codebase.
I co-authored this study of how professional developers understand unfamiliar codebases. The research and system were collaborative work. The paper appeared at ICPC 2026 and received an ACM SIGSOFT Distinguished Paper Award.
Learning from professional practice
The study drew on interviews with eight professional code auditors. We translated the findings into CodeMap, a system powered by a large language model (LLM) that combines hierarchical codebase visualisations with interactive navigation.
A developer can move from a project overview to a component and then to specific implementation details. The map keeps the repository’s structure visible while the developer explores explanations and asks follow-up questions.
Evaluation result
Among experienced developers, CodeMap reduced time spent reading LLM responses by 79%.
The published paper describes the study and evaluation. The team project page provides a demonstration of the interaction.