How it works:
The technology behind this is composed of three main parts: the fetcher, the API, and the page.
The fetcher is the GitHub-crawling robot that's responsible for fetching collaboration data for public repositories. It's written in Ruby and is basically just a set of Resque jobs that use the GitHub API to populate the database and enqueue other Resque jobs.
The API is the webapp responsible for actually computing the paths between different authors. It's a Compojure application that provides a couple of API routes. It manages to be as fast as it is by building a graph out of all the authors and projects when it starts up and then just keeping that in memory. That makes it a bona fide memory hog, but it's the only way I could get it reasonably fast without using many many terabytes of disk to store the graph's predecessor matrix.
The code for all three parts lives in the same repository; take a look if you're curious.