I usually look at each language/framework and see what problem they were designed to solve, and then implement something directly that they were designed for as my first foray. Then I will try something slightly more specific to my own needs. But normally I already have something in mind where I feel that language X will solve it better or framework Y is more suited to this problem. And even when I am just playing almost all the authors have a few examples or use cases they feel their language/framework solves really well so I will take a simple one and implement it.

At least for me, I found this keeps me honest and using the tool the way it was intended versus bending/breaking it for my needs just to say I used it. Then after I mostly understand it I can see how far I can bend it to my needs, when required.