Developed and integrated a sliding window rate-limiting mechanism for the RemoteInferenceEngine within the oumi-ai/oumi repository, focusing on optimizing API usage and preventing system overload. The solution introduced both requests-per-minute and tokens-per-minute controls, enabling more predictable and stable handling of remote inference workloads during periods of high traffic. Leveraging Python for backend development, the implementation utilized asynchronous programming techniques to enforce throttling and observability without impacting performance. This feature enhanced API stability, supported safer scaling, and improved cost control. The work emphasized robust systems design, collaborative Git-based workflows, and thorough unit testing to ensure reliability and maintainability of the new functionality.