Add a layer of abstraction to the data extraction pipeline

[BalduinLandolt]

If we want multiple data sources in the future, we should have a way of easily accommodating this. With the current architecture, that would be hard to do.
We should try to apply dependency inversion to make data extraction pipelines "plug-and-play-able".