Worked on reliability and flexibility improvements for large-scale machine learning inference engines and backends. In the ggml-org/llama.cpp and Mintplex-Labs/whisper.cpp repositories, addressed multi-threaded bugs in the clamp_f32 function, ensuring correct handling of multi-dimensional tensors and standardizing numeric semantics across projects. Leveraged C programming and multi-threading expertise to enhance production inference reliability. Later, contributed to pytorch/executorch by implementing dynamic shape support for the Arm backend, enabling flexible tensor operations and fixing resize logic for both static and dynamic shapes. Utilized Python and PyTorch skills to improve deployment flexibility and broaden model support for dynamic input scenarios on Arm hardware.