Worked on the secretflow/spu repository to deliver comprehensive Chinese-language documentation for the Policy-SGD optimizer, focusing on challenges of stochastic gradient descent in multi-party computation machine learning. Compiled and published experimental results comparing Policy-SGD with naive SGD, incorporating adaptive learning-rate scaling, learning-rate decay, and early stopping benchmarks across multiple datasets. Established a reproducibility framework to serve as a baseline for future optimizer development and to accelerate onboarding for Chinese-speaking engineers. The work emphasized technical writing, machine learning concepts, and documentation skills, with all contributions made in PO language to ensure accessibility and clarity for the target developer audience.