At one point, their search used LucidWorks Fusion [1], a commercial product based around Solr (that uses Lucene indexes under the hood) but also integrates a vector database and the like for semantic search. The linked wiki page still has Lucene-style queries.