Worked on enhancing the evaluation workflow for the lmms-eval repository by implementing new accuracy metrics for the Video-MME-v2 evaluation framework. Focused on improving metric reliability and data handling, the work involved standardizing data ingestion and correcting file path resolution for video and subtitle files using Python. These changes reduced intermittent errors and enabled more trustworthy model comparisons, supporting faster, data-driven decision making for model improvements. The updates improved traceability and production readiness through a clear commit history and targeted fixes. Core skills applied included Python programming, data processing, and video processing, contributing to a more robust evaluation pipeline.