We are thinking about building a framework that can be used for unlabeled data selection (using active learning) to reduce human labeling budgets. And hope it can be easily used as a service, currently we are adding more SOTA active learning algorithms and there is a demo for you to play with!