Ask HN: Experiment Reproducability for a production machine learning system

We are currently working on a machine learning backend and are currently looking at “experiment reproducability”, meaning the ability to go back to a particular release of the environment (code, database schema, data contained) to assess the accuracy of our model(s) at a given instant of time. I was wondering how the community deals with the above challenge. Are there tools that help with the same? A search yielded https://dvc.org/.