I was working on another project and found it easy to fine tune, but difficult to manage and improve that training set over a many versions. Would love to hear if this is a problem you've had and if you would find something like this useful.