I developed interfaces to call my company’s neural network generator directly from an application. Currently we have C/C++ and MATLAB APIs (and a shell script), and plan to add R and Python APIs soon.

Quick background: C Deeply’s neural network generator is a) parameter-free (it finds the architecture), and b) very fast (1-second computation time). It’s for dummies like myself who want lots of ML modules and don't know how to fiddle all the knobs.