Worked on the pytorch/pytorch repository to deliver backend kernel customization enhancements, focusing on foundational refactoring of CppKernelProxy and CppScheduling. Leveraged C++ and Python to introduce class-level attributes for backend-specific kernel customization, which reduced code duplication and improved type safety across kernel code paths. This approach refined the code generation process to support overridable class attributes, enabling more extensible and maintainable backend kernels. The work laid the groundwork for scalable backend extensibility and safer kernel customization, emphasizing backend development best practices. No bugs were fixed during this period, with efforts concentrated on delivering this core feature enhancement for future maintainability.