Developed extensible multidataset transfer learning support for the ecmwf/anemoi-core repository, focusing on robust data processing and model training workflows in Python. The work introduced dictionary-based data_indices loading from multidataset checkpoints, enabling flexible experimentation and seamless integration when datasets are added, removed, or swapped. Enhanced the AnemoiTrainer to validate dataset compatibility during transfer learning, reducing the risk of silent incompatibilities. Implemented a model-name-to-index mapping for accurate checkpoint loading and expanded unit tests and documentation to support these features. Collaborated with co-authors to accelerate delivery, ensuring the solution was well-tested and ready for multi-GPU validation in machine learning scenarios.