Are you interested in an ambiguous project or one where the rules are pretty well-defined? Both are useful: the latter cultivates the ability to connive solutions to clear problems (building the mechanics), and the former cultivates the higher-level ability to dissect and attack a problem.
As an example of the former, just fish around for ideas from this and other sites, or just try doing something based on your own idea.
As an example of the latter, try to build something where the protocol is well-defined. For example, build a simple web server or quantitative trading platform or web-based git client. In all cases, the protocols are fairly straightforward, and you merely have to implement them.