This project is somewhat unconventional in its approach, but that might reveal issues that are masked in typical benchmark datasets