Worked on sgl-project/sglang and zhaochenyang20/Awesome-ML-SYS-Tutorial, delivering both backend optimizations and documentation improvements. Enhanced sglang’s attention mechanism by optimizing CUDA-based caching in PyTorch, reducing redundant operations and improving throughput for attention-heavy inference. Addressed stability issues in the DeepSeek V4 FP4 indexer, resolving AttributeError and warp mask bugs to increase reliability. In Awesome-ML-SYS-Tutorial, improved multilingual documentation for the AReaL code walkthrough, adding Chinese and English versions and refining formatting for accessibility. Fixed navigation links to streamline onboarding for international developers. Demonstrated strengths in deep learning, GPU programming, technical writing, and maintaining robust, accessible open-source resources.