Interfaces for machine studying (ML), data and visualizations about fashions or information, will help practitioners construct sturdy and accountable ML methods. Regardless of their advantages, current research of ML groups and our interviews with practitioners (n=9) confirmed that ML interfaces have restricted adoption in follow. Whereas present ML interfaces are efficient for particular duties, they don’t seem to be designed to be reused, explored, and shared by a number of stakeholders in cross-functional groups. To allow evaluation and communication between totally different ML practitioners, we designed and applied Symphony, a framework for composing interactive ML interfaces with task-specific, data-driven elements that can be utilized throughout platforms resembling computational notebooks and net dashboards. We developed Symphony by way of participatory design classes with 10 groups (n=31), and talk about our findings from deploying Symphony to three manufacturing ML tasks at Apple. Symphony helped ML practitioners uncover beforehand unknown points like information duplicates and blind spots in fashions whereas enabling them to share insights with different stakeholders.