Kohan Kha contributed to the UKGovernmentBEIS/inspect_evals and VectorInstitute/vector-inference repositories by expanding embedding model support and improving inference workflows. He stabilized dependencies by pinning SWE Bench to a compatible version, addressing potential integration issues. In VectorInstitute/vector-inference, Kohan added support for bge-base-en-v1.5 and all-MiniLM-L6-v2 models, optimizing inference tasks with enhanced vLLM scripting. He also developed a Python-based demo script to showcase OpenAI API embedding generation, streamlining onboarding and testing. His work emphasized configuration management, documentation updates, and shell scripting, resulting in more robust, maintainable codebases and smoother model integration for natural language processing applications.