Francesco Milletari enhanced the PrimeIntellect-ai/prime-rl repository by delivering a robust solution for checkpoint management in Python-based backend systems. He addressed a recurring NotADirectoryError that occurred during the deletion of older checkpoints by replacing file-based removal with directory-wide cleanup using shutil.rmtree. This approach improved the reliability and safety of file system operations, particularly during long-running training runs, and reduced the need for manual intervention. Francesco’s work focused on backend development and targeted cleanup logic, resulting in safer automation and maintainability. The solution demonstrated thoughtful engineering depth, supporting future scalability and refactoring within the project’s checkpoint management workflow.