Research

Teralizer automatically transforms JUnit unit tests into property-based jqwik tests by deriving input specifications from program semantics via symbolic analysis, improving mutation scores on generated tests while surfacing practical barriers to wider applicability.

A controlled course experiment with 84 students across 14 teams shows that tying small per-sprint bonuses to automated SAST results (Bearer, Detekt, mobsfscan) significantly lowers security issue density (β = −0.396, p = 0.0342).

PASDA uses differential symbolic execution to classify behavioral equivalence between program versions, and introduces best-effort heuristics for cases where no formal proof can be found — outperforming three existing tools by 3–7% on a standard benchmark.

Through 17 interviews across 11 companies, we identify six REST and event-driven API evolution strategies and six challenges, revealing tight organizational coupling and consumer lock-in as the two core problems — and proposing automated change impact analysis and better change communication as research directions.

Doctoral symposium paper proposing a new equivalence checking approach that goes beyond accuracy — providing richer, developer-oriented descriptions of behavioral differences to better support debugging and program comprehension.

A large-scale mining study of 90,000 Maven Central projects finds that design patterns are positively correlated with software functionality and reusability but negatively correlated with understandability — consistent with smaller prior studies.