TrainLoop builds tools that let experts turn their knowledge and judgment into specialized models. Designing those tools means working in a new category, where many of the basic interactions are still open questions. This is an unusually esoteric design role. What should an end user feel when the model they work with is learning over time? How should they understand what it has learned, where it is improving, and when to trust it? How can we build systems that uncover what insights live in an expert’s head by observing the decisions, corrections, and distinctions they make in their everyday work? You’ll shape how people teach models, recognize their own judgment in the results, and guide their improvement over time. There is room here to define the interaction patterns of the category, from the first sketch through the details of the shipped interface.