Developed and integrated a new Vulkan 'where' operation for the pytorch/executorch repository, enabling conditional tensor selection within the Vulkan backend’s compute graph. This feature allows models to execute conditional data paths directly on Vulkan-supported devices, broadening deployment options for mobile and embedded environments. The implementation required expertise in GPU programming, shader development, and tensor operations, utilizing C++ and GLSL to ensure efficient backend performance. By expanding the compute graph’s flexibility, the work supports more complex model logic and accelerates feature delivery for downstream users. No major bugs were addressed during this period, with efforts focused on feature development and backend integration.