Hi Leonid,
That's a very good question. Thanks for asking. I changed the topic to reflect the new subject: SolrSherlock was originally created using Apache Solr.
That was for two very good (I thought) reasons in the beginning. First, Solr came with many plug in features I thought would be helpful. Second, the book Taming Text was just becoming available in "early access"; it is a great book from the perspective that it is a cookbook on using Solr and OpenNLP to build question answering systems.
There is, in fact, a GitHub repo for SolrSherlock.
But, as I developed code and started loading text into the system to be read and parsed, I soon realized that the architecture could be improved by separating processing agents from the indexing engine: the architecture switched from being Solr with plug in features, to Solr as index with remote agents running in different JVM instances.
I then realized that I could take advantage of some features in ElasticSearch, so the project name was changed from SolrSherlock to just OpenSherlock, leaving room for varieties of implementation details.
In a recent talk delivered in Tokyo, I describe the system in a bit more detail and explain what is behind the project. Those slides are online here:
Over time, I will be adding more details. There is now a repo at GitHub for OpenSherlock. There is a shell for the HyperMembrane component, but it is still in such a high rate of evolution (refactoring) that it putting the code online seems premature.
Cheers
Jack