Kam Salahi developed and enhanced dataset integration and evaluation workflows for the stanford-crfm/levanter and marin-community/marin repositories, focusing on robust internal supervised evaluation and expanded benchmark coverage. Using Python and YAML, Kam implemented new data ingestion pipelines, integrated datasets such as ARC, Hellaswag, OpenQA, PiQA, and Winogrande, and established reproducible download mechanisms via Hugging Face utilities. The work included extensive code linting, documentation improvements, and targeted bug fixes to ensure maintainability and reliability. These efforts improved model evaluation accuracy, accelerated experimentation, and strengthened data pipeline quality, addressing real-world machine learning operations and supporting collaborative development practices.