Worked on backend enhancements for NVIDIA/TensorRT-LLM and kvcache-ai/Mooncake, focusing on performance metrics, protocol compliance, and network configurability. Improved TensorRT-LLM by adding a maximum requests field to performance metrics and aligning usage statistics defaults with OpenAI protocol, using C++ and Python for backend and API development. For Mooncake, introduced configurable client port ranges through environment variables, enabling flexible deployment across varied network environments, and corrected documentation to ensure clarity between code and configuration. Contributions emphasized robust configuration management, precise documentation, and protocol alignment, demonstrating a methodical approach to backend development and integration using C++, Python, and Markdown.