Lei Zhou contributed to both the yhyang201/sglang and ROCm/pytorch repositories, focusing on backend and distributed systems engineering. In sglang, Lei implemented LoRA path support for chat APIs, enabling optional model customization by extending request models and functions with new parameters. This work, done in Python, enhanced the flexibility of chat workflows and laid groundwork for broader LoRA integration. In ROCm/pytorch, Lei addressed a shape-mismatch bug in the Fully Sharded Data Parallel module by introducing a tensor flattening step before copying, improving reliability for distributed training. The work demonstrated depth in API development, PyTorch, and deep learning infrastructure.