This project was done at a client site and they graciously open-sourced.
It uses word2vec vectors in its C binary representation and adds extra terms that are similar in vector space to the terms in the query.
See examples and extra documentation at its GitHub repo.