It's both easier and harder. The scale part is easy (the index is <1TB), but there's almost no training data for feedback.