Kshitij Khode developed the export of Scaled Dot-Product Attention (SDPA) to the ONNX Attention operator in the pytorch/pytorch repository, focusing on enhancing interoperability for ONNX-based workflows and production deployment. He implemented this feature using Python and PyTorch, ensuring compatibility with ONNX opset version 23. The work included comprehensive unit testing to validate the export path and maintain reliability across different deployment scenarios. By enabling SDPA to be exported directly to ONNX, Kshitij addressed a key integration challenge for deep learning models, allowing smoother inference and deployment pipelines for machine learning practitioners working with ONNX and PyTorch.