During a two-month period, Huangzhilin contributed to both the flashinfer-ai/flashinfer and kvcache-ai/sglang repositories, focusing on backend development and performance optimization. He integrated SGLang comparison into the flashinfer top-k benchmarking workflow, enhancing reporting and user-visible metrics using CUDA and Python. Huangzhilin also improved robustness by fixing chunk boundary handling in multi-CTA collaboration, expanding test coverage for variable-length data processing. In kvcache-ai/sglang, he stabilized the tuning pipeline for Fused MOE Triton by correcting ds32 configuration retrieval, reducing production risk. His work demonstrated depth in parallel computing, data analysis, and unit testing, addressing both feature delivery and critical bug fixes.