Many developers, particularly in hacking and gaming communities, use creative technogeek usernames such as z3r0c001 or B14CKH4K3R. These stylized identities make it difficult to link the same person across different platforms.

We developed GeekMAN, a systematic approach that connects technogeek usernames across forums, GitHub, and social platforms. The system translates leetspeak into readable text, splits complex handles into meaningful parts, and compares them using semantic similarity measures. GeekMAN achieved up to 86% precision on technogeek datasets, improving cross-platform linkage by 10–20 percentage points over previous methods, and is available as a publicly accessible research tool.

Developers frequently search GitHub to learn or reuse code, but existing search tools remain largely keyword-based and opaque. The question we address is: given one GitHub repository, how can we find others that are similar in purpose and functionality?

To answer this, we built MetaSim and RepoScope, two systems for exploring code ecosystems. MetaSim studies how metadata such as repository descriptions, topics, and README files define functional similarity, showing that combining these signals yields a more accurate and interpretable view of project relationships. RepoScope extends this to support "search by example repository": it enables multi-level exploration from titles and metadata to source-code embeddings, and visualizes clusters of related projects for clearer interpretation.

Thousands of GitHub repositories host malicious or dual-use code. In MAGNET, we take a software-engineering view of this malware ecosystem on the GitHub network, uncovering the technical and social mechanisms that shape how such repositories are created, forked, maintained, and how they evolve over time. This work was accepted to ASONAM 2026.