This is pretty cool. It reminded me of Hillview (https://github.com/vmware/hillview), something I worked on a few years ago while interning at VMware research (a few team members in fact used to work at Microsoft Research).

Hillview is a lot less fancy at this point, but the ambition is that it scales a lot better (that is, to way more than 500K rows. In the order of trillions).

Paper here: http://www.vldb.org/pvldb/vol12/p1442-budiu.pdf