When data scientists work on building a machine learning model, their experimentation often produces lots of metadata: metrics of models you tested, actual model files, as well as artifacts such as plots or log files. They often try different models and parameters, for example random forests of varying depth, linear models with different regularization rates, or deep learning models with different architectures trained using different learning rates.