Worked on the sktime/sktime repository to enhance the reproducibility and reliability of proximity-based estimators by addressing a critical bug in Proximity Forest and Proximity Stump. The solution involved implementing deterministic random state management by pre-generating random states for all trees, ensuring consistent results across runs. To resolve serialization issues, the approach included reusing random number generators and instantiating new transformers within distance computations, thereby avoiding closure-related pitfalls during pickling. Leveraging Python and skills in algorithm implementation, parallel processing, and machine learning, the work improved experiment reliability and deployment safety for teams relying on these estimators in production environments.