IBM architect explains how to scale AI pipelines for peak traffic WFTV