Worked on stabilizing the data preprocessing pipeline for the TabPFN repository at PriorLabs, focusing on resolving alignment issues with categorical indices. Addressed a critical bug by reverting the transformer from OrderPreservingColumnTransformer to the standard ColumnTransformer, which restored correct data alignment and improved pipeline performance. This adjustment reduced downstream errors during model training and inference, enhancing data integrity and reliability. The work aligned the pipeline with established scikit-learn practices, simplifying future maintenance and reducing technical debt. Utilized Python for implementation, applying skills in data preprocessing, machine learning, and unit testing to ensure the pipeline’s stability and maintainability throughout the process.