There's a company/open source community where they automatically parse government databases. Makes it as easy as a git clone (obviously using different syntax) for you to have a local copy - think things like holidays, weather histories, open government data (in the US).

I think journalists would love to be able to query that data via natural language. Things like "what is the most common zip code for a specific traffic violation over the past 3 years" would be super useful.

I can't remember the name but will do some searching later.