Worked on the microsoft/openvmm repository to deliver a scalable, latency-aware servicing workflow focused on NVMe keepalive integration. Developed a per-CPU selective sidecar startup mechanism during servicing restore, enabling CPUs with outstanding IO to be kernel-started while others remain sidecar-started, which reduced servicing latency and improved parallelism. Implemented robust per-CPU state management and NUMA-aware control page mappings, supporting up to 400 CPUs within a single 4 KiB page for enhanced scalability in multi-NUMA environments. Leveraged Rust for kernel development and system programming, and expanded test automation to validate per-CPU override behavior under heavy IO and high VM density scenarios.