They're using a graph database and their web of trust is still naively based on follows. But they have figured out what kind of read queries are useful for a Twitter clone, and that's worth using as a reference.