Worked on optimizing the RDMA transfer path in the kvcache-ai/Mooncake repository by implementing round-robin scheduling for slice batching across multiple queue pairs, which improved data transfer throughput and load balancing for high-concurrency workloads. The approach involved adding multi-QP spraying logic to the RdmaEndPoint::submitPostSend function using C++ and advanced RDMA techniques, addressing both performance and reliability. Collaborated closely with another contributor to ensure code quality and effective integration. This work demonstrated strong skills in C++, RDMA, and system programming, delivering tangible improvements to network programming efficiency and robustness within a complex distributed system environment.