1. feed it a ton of documents
2. either procedurally (ideally) and/or manually (a la wikipedia but with a fast interface) link between them.
3. Also create entities, so like, maybe Bernie Madoff is an entity, and there are companies that are owned by him, those companies might have phone numbers. (totally bullshitting this example so please ignore legal, investigative, factual nonsensicality)
4. You keep putting all the docs you run across into the system as you go and one day you put in a doc from an investigation of a shady shell corp. One of the phone numbers listed is recorded as a Madoff company's number and you see that as you load in the doc. Either it's relevant for you or the next person to go look for info about Madoff would see it. That connection might never get made without the built up knowledge base and auto-associative kind of features.
I always looked at it like a super easy to use wiki or document database that had investigatory tools built on top of it. The search was really well done and following relationships was easy. Granted I was comparing the experience to other gov't software, the fact that it worked almost all the time and didn't drive me to script my interactions with it pretty much alone made it my favorite. So Palantir might objectively suck and I was just so mired in terrible that it looked relatively stellar, but I remember it as useful and easy to use.
As a customer of Palantir, am I responsible for creating the document parsers that understand the relationships between entities or is their software able to recognize data types in the inputs automatically? Is the software generic or customized for each customer that comes in the door?
[0] I know it's an odd description but I think it puts the right idea in your head. Look at the little desperado/mystery man icon: https://www.palantir.com/wp-assets/wp-content/static/techblo...
What's the general gist of what Palantir's software does? What's the input and what's the output?