Worked on enhancing the Vertex AI integration within the menloresearch/litellm repository, focusing on improving reliability for Gemini model workflows. Addressed HTTP response handling by implementing logic to treat HTTP 201 Created as a successful outcome for both asynchronous and synchronous API calls, reducing false negatives in production. Expanded test coverage using Python, emphasizing API integration testing, error handling, and mocking to ensure robust detection of non-200/201 responses and proper raising of VertexAIError. The work prioritized maintainability and risk reduction, resulting in a more resilient integration that supports safer deployments and clearer diagnostics for teams leveraging Vertex AI through this library.