Worked on enhancing model versatility and stability across the UKGovernmentBEIS/inspect_evals and VectorInstitute/vector-inference repositories. Delivered new embedding model support and improved inference handling by integrating bge-base-en-v1.5 and all-MiniLM-L6-v2 models, while optimizing vLLM-based task execution. Addressed dependency management by pinning SWE Bench to a stable version, ensuring compatibility and reducing integration issues. Developed a Python script to demonstrate OpenAI embedding API usage, streamlining onboarding and testing. Maintained strong documentation and configuration practices, updating environment variable handling and reflecting new model variants. Utilized Python, Shell scripting, and TOML to deliver maintainable, well-documented solutions focused on reliability and usability.