Devang worked on the pytorch/ao repository, where he developed ArmInductorQuantizer support for Pt2e quantization on ARM architectures. His work focused on enabling quantization for conv2d and linear layers, with comprehensive unit tests to ensure correctness and reliability. Using Python and leveraging PyTorch’s quantization and machine learning capabilities, Devang addressed CI issues and improved code style adherence, which enhanced automated validation processes. He also expanded test coverage for various quantization scenarios and laid the groundwork for broader ARM hardware deployment. The depth of his contribution was in building robust, tested features that facilitate efficient model performance on ARM platforms.