Worked on the google/adk-python repository to enhance Vertex AI integration by addressing a critical reliability issue. Developed a Python utility that strips unsupported part_metadata fields from event payloads before requests reach the Vertex AI Agent Engine Sessions API, preventing 400 INVALID_ARGUMENT errors caused by Gemini-specific metadata. Employed backend development and API integration skills to ensure the solution maintained data integrity, preserving part text while removing problematic metadata. Added comprehensive unit and regression tests using Pytest to validate the fix across both content and raw_event payloads. Collaborated on code review, adhering to formatting standards and improving overall code quality and stability.