Xuan worked on the BerriAI/litellm repository, focusing on improving Vertex AI integration by addressing a persistent API compliance issue. Using Python and leveraging skills in API development and unit testing, Xuan separated Tool objects by type—FunctionDeclaration, GoogleSearch, and CodeExecution—to align with Vertex AI’s API specifications and eliminate INVALID_ARGUMENT errors. This modular approach not only reduced integration friction but also enhanced maintainability for future expansions. Xuan updated the test suite to verify the new object separation, ensuring robust coverage and smoother automated workflows. The work demonstrated careful attention to architectural clarity and practical problem-solving within a complex API environment.