Over a three-month period, contributed to NVIDIA/NeMo-Skills and NVIDIA-NeMo/Eval by building and refining data pipelines, evaluation workflows, and command-line tooling. Enhanced the AA-LCR data pipeline with robust error handling and automated dataset acquisition from Hugging Face Hub, improving reliability and maintainability using Python and structured logging. Developed structured output support and a dedicated metrics class for the HLE judge, enabling granular evaluation and reproducibility in machine learning assessments. Improved summarization accuracy by filtering output files and introduced intuitive task-name-based CLI overrides, streamlining configuration management. The work emphasized data processing, API integration, and Python scripting to support scalable model development.