We could do something where we "compress" dictionary by collapsing keys where common prefix points to unique terminal station or something like that, but probably easier to just dump all the Trips in SQLite and query on demand to avoid keeping all this junk in memory.