Worked on the pytorch-labs/monarch repository to deliver Elastic Fabric Adapter (EFA) support within the RDMA subsystem, enabling high-performance networking for scalable HPC workloads. Developed auto-detection of EFA devices and implemented EFA-specific connection handling in C, while introducing new data transfer paths for non-tensor buffer types such as bytearray and memoryview. Maintained compatibility with existing RDMA (mlx5) paths by isolating EFA logic and using an explicit is_efa flag. Static linking of libefa.a was adopted to prevent runtime conflicts. Comprehensive end-to-end validation on EFA-enabled AWS hardware ensured robust performance and stability across multiple buffer types and sizes.