Worked on the NVIDIA/NeMo-Skills repository to enhance the LLM-based evaluation pipeline by implementing LLM-as-a-judge functionality. Developed new CLI arguments in Python to allow users to control judge generation type and module, improving flexibility in evaluation workflows. Updated Markdown documentation to guide users in evaluating natural language math benchmarks, ensuring clarity and usability. Added targeted test cases to validate the LLM-as-a-judge feature with API-based servers, emphasizing robust test coverage. Focused on CLI development, documentation, and LLM evaluation, the work addressed both feature delivery and quality assurance, resulting in a more adaptable and well-documented evaluation pipeline for the project.