Harsh Sharma developed and integrated the GroundCocoa benchmark task into both the red-hat-data-services/lm-evaluation-harness and swiss-ai/lm-evaluation-harness repositories, focusing on evaluating compositional and conditional reasoning in language models for flight booking scenarios. He designed new YAML-based task configurations and implemented Python processing utilities to handle dataset documents, ensuring the benchmarks were robust and extensible. Harsh also updated documentation in Markdown to improve onboarding and configuration clarity for contributors. His work demonstrated depth in benchmark development, data processing, and machine learning evaluation, laying a foundation for future domain-specific assessments and enhancing the codebase’s readiness for further scaling and adoption.