Worked on the rapidsai/raft repository to deliver a targeted performance optimization for the Workspace Resource Factory, focusing on reducing allocation latency in CUDA contexts. The solution conditionally bypassed the default_allocation_limit logic, thereby avoiding unnecessary cudaMemGetInfo() calls when an explicit allocation_limit was set. This approach streamlined the allocation path, resulting in improved responsiveness and scalability for workloads with high concurrency. The work was implemented in C++ with CUDA, emphasizing performance optimization and disciplined code review. The contribution addressed a specific bottleneck, enhancing throughput for allocation-heavy operations while maintaining code clarity and process rigor throughout the development cycle.