Worked on the pytorch/pytorch repository to improve the reliability of CI pipelines by addressing environment-specific patching for Numba’s CUDA-13 support. Focused on restricting the application of a CUDA-related patch exclusively to CUDA environments, thereby preventing unintended issues in ROCm environments and reducing test flakiness. This approach established a clearer separation between CUDA and ROCm workflows, supporting more consistent cross-ecosystem development. Utilized Python and Bash scripting to implement and validate these changes within the CI/CD process. The work emphasized robust scripting and patch governance, ensuring that future CUDA-related updates can be managed with greater precision and reliability.