You should read my dissertation here: http://lenary.co.uk/publications/dissertation/ it gives enough of an overview that you should see what is doable and what isn't. Of course the writing isn't super great, sorry about that.
Then, of course, realise that this is about 1/3 finished, but pretty much abandoned right now. Here are some jumping off points to taking it forwards:
Some ideas for jumping off points:
- look at the effects system in Idris, how this might work with the models of concurrency seen in erlang
- how do the different trade-offs that idris makes affect how good this will be at distribution. Concurrency is a slightly different problem, where I assumed a lack of failures (ie if one thing failed, it's fine for everything to fail).
- Better types for behaviours should be doable. I touched on this in my dissertation.
- There's a haskell library for generating core erlang. That could be a better backend than escripts.