Worked on the xlang-ai/OSWorld repository to enhance reliability and scalability for large-payload workflows, focusing on backend development and robust error handling. Addressed issues with oversized requests by implementing comprehensive detection and retry logic for HTTP 413 and related errors, including automatic reduction of image counts and seamless retries across both primary and backup API paths. Utilized Python and JSON to extend error coverage, ensuring safer and more stable evaluations. The improvements reduced evaluation failures and enabled higher throughput for large submissions, with clear documentation provided through example JSON workflows. The work emphasized API integration and resilient backend infrastructure.