Worked on the sktime/sktime repository to address compatibility issues in the ChronosForecaster module following API changes in Transformers 5.X. Focused on stabilizing the forecasting suite by standardizing encoder and decoder initialization using T5Stack and updating input embedding handling, as well as implementing a custom _tie_weights override to maintain correct weight tying after initialization. Added targeted end-to-end tests in Python to verify instantiation, fitting, and prediction workflows, which improved CI reliability and reduced maintenance overhead for users. This work leveraged skills in Python, machine learning, and software development to ensure cross-version interoperability and address failing tests in the codebase.