Aash Mohammad developed privacy-preserving machine learning capabilities for the APPFL/APPFL repository, focusing on integrating differential privacy into federated learning workflows. He implemented Opacus-based training, enabling configurable privacy options and Gaussian mechanisms to enhance data protection during model training. His work included updating a ResNet variant for compatibility with Opacus and extending the VanillaTrainer to support differential privacy workflows. Using Python and YAML for configuration management, Aash also updated a federated learning notebook to demonstrate privacy-preserving experiments. The work addressed the need for robust privacy controls in federated settings, reflecting a deep understanding of both differential privacy and ML engineering.