In February 2025, Liang Songxiang enhanced the XENONnT/straxen repository by upgrading its TensorFlow/Keras model loading protocol to support Keras 3, focusing on improving machine learning model deployment reliability. Using Python and configuration management skills, Liang removed the legacy tf:// protocol, registered classes from keras.tar.gz, and addressed incompatibilities with NumPy 2. These changes streamlined the packaging process, reduced runtime errors, and broadened compatibility for modern ML workloads. Liang also tuned the posrec plugin to align with updated dependencies, resulting in a more maintainable codebase and enabling faster integration of new machine learning models into straxen’s data processing workflows.