I think this is an important stepping stone towards automating some development tasks.

But the execution went wrong... They used raw data from repositories for training, without any kind of pre-processing.