Worked across Intel-tensorflow/tensorflow, google-ai-edge/LiteRT, and openxla/xla to deliver scalable tensor operations and improve system robustness. Enabled 64-bit indexing for large-scale tensor computations, allowing support for larger datasets and improved performance. Addressed stability by fixing bounds checks, use-after-free errors, and hardening Windows DLL loading. Enhanced TPU utilities with robust protocol buffer handling to prevent serialization issues. Improved shape and element count calculations, added overflow prevention for GPU weight layouts, and strengthened input validation. Leveraged C++, Python, and CUDA to optimize kernel development, error handling, and build automation, focusing on security, maintainability, and correctness across multiple repositories.