Worked on stabilizing CUDA sample code in the NVIDIA/CUDALibrarySamples repository, focusing on improving the reliability of sparse-matrix operations. Addressed a critical bug in the SpMM CSR Operation sample by correcting the NVRTC compilation process, specifically replacing incorrect LTIOIR retrieval with proper NVVM usage and ensuring accurate architecture string formatting. This fix removed build-time blockers and enabled correct kernel compilation for SpMM workloads across targeted GPUs. The work demonstrated proficiency in CUDA, GPU computing, and NVRTC, resulting in more dependable sample code and reducing setup time for developers evaluating sparse matrix operations in C within the repository’s codebase.