Worked on enhancing the stability of the PriorLabs/tabpfn-extensions repository by addressing a critical bug in the TunedTabPFNBase class. Focused on preventing out-of-bounds random seed generation, the solution involved constraining random integer values to the int32 range during data splitting and model initialization. This update, implemented in Python, improved the reliability of experiment reproducibility and reduced runtime errors related to ValueError exceptions. The work demonstrated proficiency in deep learning, hyperparameter optimization, and debugging, with careful attention to random number generation and seed management. The contribution strengthened deployment readiness by ensuring safer and more predictable model workflows.