Developed the Kubernetes Scheduler Resource Customization feature for the pytorch/torchx repository, enabling users to override CPU and memory overhead as well as AWS EFA device count within the TorchX Kubernetes scheduler. This enhancement addressed the need for precise resource management and improved performance for containerized HPC and AI workloads leveraging EFA. The work involved Python development and deep integration with Kubernetes, focusing on scheduler customization and resource allocation. Collaboration with infrastructure and CI teams ensured robust validation of customization scenarios, resulting in a scalable and reliable solution. The feature was merged following a focused review process, with no major bugs reported.