Worked on the sglang repositories to enhance model optimization workflows and developer experience through targeted feature development and documentation. Focused on improving CI reliability by adding comprehensive math evaluation tests and updating score thresholds, while also strengthening nightly test coverage using Python and unittest. Delivered detailed documentation for DeepSeek model optimizations, including MLA and tensor parallelism, to guide users in maximizing inference efficiency. Integrated API and LLM support for DeepSeek V3.1 and Kimi-K2, updating usage guides and correcting tool-call ID formatting in JSON to ensure reliable tool invocation. These contributions improved onboarding, reduced misusage risk, and accelerated feature adoption.