What helps me with this approach is to pick a small module, and then start writing a README for it as if the module already existed.
By pretending that the module is already done you get to use it before it exists (which is a lot of fun :). In the README, write your instructions in such a way that they are testable (python doctests make this easy, I assume you can do something similar with ruby).
When you're done, it is much easier to start on the actual module itself because you've already figured out how that module interacts with the rest of the system, and you have a set of tests to verify it.
By pretending that the module is already done you get to use it before it exists (which is a lot of fun :). In the README, write your instructions in such a way that they are testable (python doctests make this easy, I assume you can do something similar with ruby).
When you're done, it is much easier to start on the actual module itself because you've already figured out how that module interacts with the rest of the system, and you have a set of tests to verify it.