Yeap, the problem was the VM. And we are going to opensource most of it.

We tried to find something that was fast, cheap to scale (in terms of servers) and easy to extend but we couldn't find any other solutions. The problem is having everything in memory and doing correlations really fast.