Worked on the facebookincubator/velox repository to refactor cuDF operators, unifying them under a new CudfOperatorBase class to streamline code organization and improve maintainability. Leveraged C++ and object-oriented design principles to reduce duplication and lay the foundation for consistent operator architecture. Integrated NVTX profiling into the operator execution paths, enabling detailed performance monitoring and facilitating data-driven optimizations. This approach supports faster diagnosis of performance issues and aligns with Velox’s architectural goals for extensibility. The work demonstrated skills in code refactoring, performance optimization, and software architecture, resulting in a more maintainable and extensible codebase without introducing new bugs.