As an intermediate step lets build a super duper simple query language against both Wikidata's SQL and Titan's Gremlin interfaces. Daniel will propose a syntax and implement it against SQL and the wikidata-gremlin team will implement it against Titan. I'm not being prescriptive on how - Gremlin plugin, MediaWiki plugin, third service, quantum computer, mechanical computer, whatever.
I would actually argue for starting with a simple subset of whatever we plan to use in the longer term (MQL?), and then expand from there. Lets not expose <n> different interfaces for the same thing just because we can. Doing so would have
a) a cost on our development due to the extra effort to build and maintain those interfaces, and
b) a cost on users by confusing them with several options.
Lets perhaps resolve T85181 before we go ahead with this?