To be honest, I would lean toward making small projects for that goal, but that might be my bias. Taking things from zero to one is a much different endeavor than fixing stuff or adding features in an existing project, though the benefit of the latter is that you get to see how other people are handling certain problems and interact with those patterns while trying to do whatever you're doing in the context of them. The skills will transfer either way, but the design of a feature can be a lot different from the design of a standalone application