Just seriously took a look at the Real Time Composition Library

Interesting, I think sometimes using other peoples patches without changing them is cheating (your own intellectual development maybe), it's like opening Logic and pulling up a preset and using it straight out of the box. I'm not sure why you would want to use someone else's interpretation (unless it fitted your musical purposes perfectly) as i find max a way to free yourself from the normal constraints/limitations of someone else's logic.

I've gone down the route of trying to build a small composition library for myself, not as complex as essl's as it doesn't need to be for my needs, I've tried to automate a lot of the processes I would usually do in a DAW, like triggering chords and playing scales. I've learned from a book that comes with pretty much all the devices pre built and ready to use, but instead of using them straight out of the box I've gone down the route of going through all the exercises and trying to understand what is happening inside to give the results so I can replicate them myself.

If you were imagining a set of composition tools that are exactly like Essl's then I reckon you should take inspiration, note flaws, adapt and create something better with more or less features using your own logic. You'll learn a heck of a lot and feel a million times better when using your own creation compared to using someone else's imo.

Everything's been done before, but no one's done it the way you would yet!

If you want inspiration of where to go instead of down this path I had a few ideas about making a perceptron that you could guide to certain chords or combinations of rhythms. Could be an interesting avenue to explore, or using neural nets to analyse real time musical input, store as bias towards certain pitch/duration/velocity values and respond to certain stimulus. Imitating someone's musical preference by machine learning in real time seems far more interesting to me than trying to imitate a set of abstractions that have been done before.

You could even explore meta tagging interesting combinations, like tagging certain results with the words moody, jazzy, happy, sad etc, then make a program that allows you to type in real world descriptive emotional words and automatically composes to that feel (This also allows you to empirically analyse other peoples subjective preferences from an objective point of view). Not to mention the usual conditioning of NN's, rewarding it for good behaviour, punishing it for bad behaviour so it doesn't crop up again. I think that's a much more relevant avenue to explore in this day and age personally. (I only mention this as you said you're a C++ head in another thread, so i thought you'd understand this mentalness!)