Developed a core feature optimization for the kvcache-ai/Mooncake repository, focusing on buffer size allocation within the P2PProxy component. This work targeted enhanced memory management and performance across both CPU and CUDA environments, addressing resource utilization and throughput for distributed peer-to-peer caching workloads. The approach involved tuning memory allocators and optimizing cross-environment performance, resulting in reduced latency and improved scalability of the caching layer. Implemented entirely in C++ with CUDA programming, the changes were delivered as a maintainable, traceable commit. No bug fixes were recorded during this period, with efforts concentrated on robust feature delivery and code maintainability.