refinery. The data scientist's open-source choice to scale, assess and maintain natural language data. Treat training data like a software artifact.