Worked on backend stability and reliability across multiple GPU programming repositories, including JuliaGPU/pocl, ggml-org/llama.cpp, and Mintplex-Labs/whisper.cpp. Focused on resolving critical bugs in C and C++ codebases, addressing header file encoding, kernel argument type handling, and memory alignment issues. Improved header hygiene and cross-header consistency in JuliaGPU/pocl to prevent future build problems. In llama.cpp and whisper.cpp, corrected OpenCL kernel argument casting and buffer alignment calculations, eliminating segmentation faults and allocation errors. These targeted bug fixes enhanced cross-device portability and runtime stability, supporting production workloads with predictable performance through careful low-level programming and memory management practices.