During May 2025, Brian Dhu expanded distributed capabilities in the facebookresearch/param repository by implementing object-based collective communication for PyTorch. He developed support for all_gather_object and broadcast_object_list operations in the distributed backend, enabling more flexible training patterns and efficient memory and bandwidth usage. His work involved integrating backend utilities, designing preparation functions, and refining tensor allocation logic to accommodate object-based operations. Using Python and C++ within a distributed systems context, Brian ensured traceable, well-structured commits and maintained system stability. This feature-rich contribution laid the foundation for future performance optimizations and addressed evolving needs in large-scale PyTorch distributed training.