Worked on the Intel-tensorflow/tensorflow repository to enhance GPU kernel safety and scalability for large workloads. Focused on enabling 64-bit work-element support across key operations by introducing new GPU launch configurations and updating kernel logic to use overflow-safe arithmetic. Replaced deprecated grid iterators with 64-bit-capable variants and implemented robust error handling using absl::StatusOr to prevent crashes during kernel execution. Leveraged C++, CUDA, and numerical methods to ensure stable and efficient training and inference on GPUs. These improvements reduced runtime failures and prepared the codebase for future performance enhancements, demonstrating depth in GPU programming and kernel engineering.