During May 2025, Zhangrui focused on enhancing the Shubhamsaboo/llm4ad repository by refactoring the profiler’s state management system. He transitioned sample counts and generation counters from class-level to instance-level attributes, improving encapsulation and testability. This change streamlined state management and made the codebase more scalable for future development. Additionally, Zhangrui removed the gym dependency, reducing external requirements and simplifying deployment. His work centered on Python, leveraging skills in class design, code refactoring, and dependency management. The depth of these changes addressed maintainability and deployment challenges, resulting in a more robust and modular profiling workflow for the project.