Worked on enhancing distributed tensor operations in the pytorch/xla repository, focusing on DTensor and XLAShardedTensor integration for scalable training. Developed features to improve mesh conversion reliability and sharding compatibility, including translating sharding information into DTensorSpec and expanding test coverage for mesh conversions across various configurations. Implemented asynchronous redistribution support in XLAShardedTensor, with comprehensive tests covering tensor shapes, dtypes, and placement strategies. Refactored the codebase to align XLAShardedTensor with the DTensor API, supporting gradient propagation and maintainability. Utilized Python, PyTorch, and XLA, emphasizing robust testing and refactoring to support distributed systems and future extensibility.