The idea is that you'd use this via the python shell (and soon, from your R code or Javascript code), and not through the website. Looks like we need to make that clearer. The website stack is Postgres+Django+Python.
How interconnected is your data? Did you consider a graph db?
It's highly interconnected. For example, you can make hops like USCounty--USState--Senator--Vote on economic stimulus bill.
The data backend is indeed NoSQL, but we didn't choose a graph db because there are no really good graph solutions that are easily parallelized.