Worked on integrating the Korean Benchmark for Legal Language Understanding (KBL) dataset into the lm-evaluation-harness repositories for both red-hat-data-services and swiss-ai, focusing on expanding evaluation capabilities for Korean legal NLP models. Employed a configuration-driven approach using YAML to support knowledge-based questions, reasoning tasks, and bar exam simulations, enabling scalable and flexible benchmarking. The work included comprehensive dataset integration, cleanup of legacy retrieval-augmented generation (RAG) configurations, and improvements to code maintainability. Emphasized configuration management, dataset integration, and machine learning evaluation, establishing a robust baseline for Korean legal language tasks without introducing new bugs during the development period.