Developed a dynamic CUDA version compatibility layer for the intel/onnxruntime repository, focusing on robust support for CUDA 12, CUDA 13, and future releases. The work centered on replacing hardcoded CUDA version logic with dynamic detection in Python, updating both initialization and build scripts to ensure correct library loading at runtime. Enhanced the build and runtime processes by introducing flexible library path management, version validation, and compatibility checks with PyTorch, while aligning packaging for TensorRT and CUDA 12+. The solution was validated end-to-end, confirming backward and forward compatibility, and demonstrated strong skills in CUDA, Python development, and library management.