Worked on the microsoft/magentic-ui repository to deliver Approval Guard compatibility with the Gemini OAI API, focusing on enhancing the accuracy and clarity of human-approval gating. Refactored the approval guard using Python, integrating API endpoints and updating system messages and prompt templates to define approval criteria more explicitly. This approach reduced ambiguity and minimized the risk of unintended auto-approvals, aligning the codebase with Gemini API requirements. Applied skills in API integration, code refactoring, and prompt engineering to ensure seamless interaction between the approval system and the new API, resulting in a more robust and maintainable approval workflow for the project.