Sharipov worked on the sarapapi/hearing2translate repository, delivering a device-agnostic tensor conversion feature that removed CUDA hardcoding and enabled compatibility with both CPU and non-CUDA GPUs. Using Python and PyTorch, Sharipov refactored the data processing pipeline to dynamically select the appropriate hardware device for tensor operations, addressing potential misconfigurations and performance issues. This update broadened deployment options and laid the foundation for future multi-device support. The work demonstrated a focused approach to improving hardware flexibility in machine learning workflows, with depth in understanding device management and practical application of PyTorch’s tensor handling capabilities within a real-world translation pipeline.