Avem developed enhancements to the LLM-based evaluation pipeline in the NVIDIA/NeMo-Skills repository, focusing on enabling LLM-as-a-judge functionality for natural language math benchmarks. Using Python and Markdown, Avem introduced new CLI arguments to control judge generation type and module, allowing users to customize evaluation workflows. The work included updating documentation to clarify evaluation procedures and creating targeted test cases to validate the new LLM-as-a-judge feature with API-based servers. Avem’s contributions demonstrated depth in CLI development, testing, and LLM evaluation, resulting in a more flexible and robust evaluation pipeline that supports advanced benchmarking scenarios for language model assessment.