Worked on the sbintuitions/flexeval repository to enhance the reliability of automated evaluation pipelines by addressing a critical issue in ROUGE metric computation. Implemented a targeted fix in Python to handle edge cases where model outputs are empty or contain only punctuation, which previously caused ValueError exceptions during evaluation. This update ensures that metric reporting remains consistent and stable, reducing runtime errors that could disrupt downstream decision-making processes. The work involved careful data processing and integration of machine learning evaluation logic, resulting in a more robust evaluation framework that supports smoother experimentation and more dependable results for the flexeval project.