Worked on DeepTrackAI/DeepTrack2 to enhance stability and reproducibility for data science workflows, focusing on test infrastructure and notebook management. Addressed reliability issues in the Python-based test suite by reverting a transformation method, restoring deterministic behavior and improving CI outcomes. In a separate effort, maintained compatibility for deep learning and image processing research by reverting a Jupyter notebook rename, ensuring downstream scripts continued to function without disruption. Emphasized clean, auditable changes with minimal workflow interruption. The work demonstrated a methodical approach to maintaining robust machine learning pipelines, leveraging skills in Python, Jupyter, and unit testing to support ongoing research needs.