Maybe they want to recycle complex analytic queries and visualization tools for rdbms-databases, while still enjoying the advantages of mongodb. They mention this in the first paragraph.
"This way, you can run complex analytic queries, join different collections, and benefit from SQL data visualization tools."
"In the following, we show a rather simple query, but complex queries that involve sub-selects, SQL window functions, and collection joins are also possible. In fact, you can even join a MongoDB collection with a PostgreSQL table."
"And the really exciting part for us is that CitusDB can easily enable analytic queries for sharded MongoDB installations, as it is built on PostgreSQL and can already distribute SQL queries."
This company builds CitusDB, "a distributed database that lets you run SQL queries over very large data sets. Designed for analytical queries, CitusDB enables real-time responsiveness." This apparently allows their software to work with MongoDB.
Those slides gave me a headache, so I'll just point out for the benefit of everyone that there is a README, and that it's some form of replication via AMQP queues.
Is there some sort of advantage to this that I'm missing? It just seems like reinventing the wheel.