Worked on the OpenPipe/ART repository to enhance preprocessing and evaluation pipelines for AI development workflows. Developed supervised fine-tuning preprocessing for trajectory tokenization by introducing the SFTBatch class and integrating the unsloth-zoo library, improving response handling and maintainability. Extended the RULER evaluation component to support tool definitions, enabling tool-aware scoring and increasing evaluation fidelity. Focused on code quality by applying Ruff lint fixes and reorganizing imports with added documentation for clarity. Leveraged Python, asynchronous programming, and data processing skills to deliver more reliable SFT experimentation and faster iteration, establishing a foundation for scalable machine learning deployment and robust evaluation.