To build reliable ML systems, model developers need access to high-quality datasets. However, there are few resources available that guide model developers through best practices for collecting training datasets. To address this gap, Meta gas released Mephisto a crowdsourcing platform that helps users share, reuse, and extend training data collection methodologies. In addition to helping users tweak existing blueprints for data collection and annotation, Mephisto also includes tools to launch, monitor, and review data labeling workflows and experiments.