Worked on edge AI runtimes in the google-ai-edge/LiteRT and Intel-tensorflow/tensorflow repositories, focusing on stability, memory management, and build-time flexibility. Refactored LiteRtRuntimeCApiStruct initialization and CPU accelerator registration to use static const initializers and conditional logic, improving memory efficiency across build configurations. Addressed integer overflow risks in TensorFlow Lite convolution operations by adding im2col size safeguards and correcting output shape calculations, enhancing reliability for machine learning workloads. Delivered a Windows-specific fix by switching LiteRT’s weight loader to memory-mapped files with enforced 64-byte alignment, leveraging the Windows API to resolve crashes and improve cross-platform deployment stability. Used C++, TensorFlow, and system programming.