2025
CodeMap
Human-AI Collaboration for Code Understanding
TypeScript · React · LLMs · OpenAI · Data Visualization · Human-AI Collaboration · Research
Description
CodeMap is a research project focused on helping developers understand unfamiliar codebases through interactive visualizations and AI-assisted exploration.
I contributed to the development of this project while working with Jie Gao. It explores how developers build mental models of complex software systems and how AI can support that process without replacing human understanding.
Rather than relying solely on conversational interactions with LLMs, CodeMap combines visual representations, hierarchical exploration, and contextual AI assistance to help developers navigate large codebases more effectively.
Motivation
Understanding an unfamiliar codebase is often one of the most difficult parts of software development. Important architectural decisions, relationships between components, and project-specific knowledge are frequently scattered across files, documentation, and team discussions.
This project explores how interactive visualizations and AI-assisted workflows can support developers in building a clearer understanding of software systems while maintaining their own reasoning and judgment.
Features
- interactive codebase visualizations
- hierarchical exploration across multiple abstraction levels
- context-aware AI-assisted code understanding
- visual navigation of software relationships and structures
- support for understanding unfamiliar codebases
- research-driven design and evaluation
Research Context
CodeMap was developed as part of research on human-AI collaboration in software engineering.
The project investigates how developers interact with AI systems while performing code comprehension tasks and explores ways of combining visual representations with LLM-assisted workflows. The resulting system was evaluated through user studies involving both experienced and novice developers.
Related publication:
Understanding Codebase like a Professional! Human-AI Collaboration for Code Comprehension
ICPC 2026 · ACM SIGSOFT Distinguished Paper Award
What I Learned
Working on CodeMap gave me deeper exposure to software visualization, human-computer interaction, and the challenges involved in designing AI-assisted tools that support human reasoning rather than simply generating answers.
It also changed how I think about code comprehension itself. Understanding software is often less about reading individual files and more about building a mental model of how different parts of a system relate to one another.