An implementation of Unstructured QA: https://arxiv.org/pdf/1608.07905.pdf (+tranfer learning for in-domain dataset)
While we don't release the the in domain dataset that we intend to fine tune our model on, our implementation works just as well on SQuAD. Most of the code is documented and I hope is easy to read as well.
We expect the inputs as pre-processed by this repo: https://github.com/MurtyShikhar/Question-Answering