Worked on stabilizing GPU memory management for the apache/sedona-db project by upgrading RAPIDS RMM to version 25.12 and applying a targeted patch to resolve a race condition. This effort focused on improving the reliability and performance of GPU-accelerated spatial computations, particularly under heavy workloads. Using C++ and CMake, addressed a critical memory management bug that previously led to inconsistent production behavior and GPU-related failures. The changes resulted in reduced memory-related errors and enabled smoother, higher-throughput production runs. The work demonstrated depth in GPU programming and memory management, contributing to more consistent and reliable operations for spatial data processing workloads.