Worked on the mfem/mfem repository to deliver a GPU-accelerated optimization for the SparseIJToCSR_DG routine, targeting LOR assembly workflows. Refactored C++ code to improve memory handling, data access patterns, and per-row nonzero element calculation, enabling faster and more reliable GPU execution using the Finite Element Method. Addressed stability by debugging GPU execution paths and later implemented defensive programming practices, such as null-pointer safeguards in marker handling, to prevent runtime errors during integration workflows. Focused on performance optimization and pointer safety, the work enhanced simulation throughput potential and improved the reliability of large-scale, marker-based integration in scientific computing contexts.