[cckolon]

This is a really cool idea. By “different models” do you mean models fine tuned on different sources? How would you decide how to classify chunks?