This would make a nice heuristic for autocompletion too.

Related would be an SQLalchemy database structure crawler that would suggest appropriate mongodb collection aggregation, based on the frequency of columns, keys, etc.