Indeed. I think too many problems in our industry are seen as unapproachable. There's a lot of people working on postgres, me included, who did not have any sort of deep background in databases before. You start working on smaller things (I started making int -> text conversion faster), review other people's patches, start to develop new features, ... Gradually that gives you a more and more knowledge in database architecture. And you read a few good papers here and there.

Obviously that approach doesn't really lend itself to writing something like postgres from scratch - but realistically that's not something you're going to do on your own anyway. And if you do start a new project you don't set out to do something absolutely complete, but build it iteratively. With more domain knowledge, you're more likely to get the architecture halfway right initially, but in either case you're going to have to redesign and redesign and redesign.