Worked on the UKGovernmentBEIS/inspect_ai and UKGovernmentBEIS/inspect_evals repositories, delivering a comprehensive overhaul of BBH evaluation subset processing to ensure correctness across all benchmark subsets. Applied data engineering and refactoring skills to address issues in dataset construction, prompt management, and solver or scorer selection, resulting in more reliable and maintainable evaluation workflows. Upgraded the beautifulsoup4 dependency to maintain compatibility and runtime stability without requiring code changes. Strengthened testing coverage and code organization using Python, which reduced manual debugging and established a foundation for scalable future enhancements. The work improved workflow stability and accelerated future feature development across both repositories.