During January 2026, Andrew William enhanced the OpenPipe/ART repository by developing features focused on AI development and data processing using Python. He implemented supervised fine-tuning preprocessing for trajectory tokenization, introducing the SFTBatch class and integrating unsloth-zoo to improve response handling and streamline batch management. Andrew also extended the RULER evaluation pipeline to support tool definitions, enabling tool-aware scoring for more accurate model assessment. His work included asynchronous programming techniques and code-quality improvements such as Ruff lint fixes and import-order cleanup, which improved maintainability. These contributions laid a foundation for more reliable experimentation and scalable deployment within the project.