Worked on enhancing the HuggingFace.js repository by integrating WBench support into the evaluation framework, focusing on benchmarking interactive video world models. Leveraged TypeScript and full stack development skills to register WBench in the evaluation workflow, enabling comprehensive multi-turn benchmarks across five dimensions, twenty-two metrics, and nearly three hundred interaction cases. The technical approach centered on safe, registry-only changes without affecting runtime logic, authentication, or data paths, which streamlined both maintenance and deployment. Prepared the WBench dataset for seamless integration with Hub Evaluation Results, allowing benchmark outcomes to be surfaced in dashboards and comparisons for improved model evaluation transparency.