Thanks to kvale on Perlmonks, I found a problem with my inline neural network module. No matter how hard I tried, I couldn't get the network over 86% accuracy. kvale looked at it and found that I was training it with conflicting data (duh!). Now that I have that worked out, I hope to finish the documentation and have it "newbie friendly" so anyone can start playing with neural networks.
It has a Perl interface, but the guts are in C for performance reasons. That means there are a few people who won't be able to play with it, but it's too slow if I write in pure Perl.
My next thought is to write a genetic "neural net" designer that hooks into the module. That might be a wee bit much for me, though. This is pretty new territory for me.