While I was at Yale, we used neural network simulators to solve (or attempt to solve) many of our pattern recognition problems. We worked with a number of classes of neural networks, but they shared one thing in common - you could twiddle their innards while they were running. Being able to watch and interact with the nets while they were running made it far easier for us to understand why they did what they did.

All our simulators were written in Pascal and ran on a VAX Microstation. I was the only programmer in the group, and was also the system administrator for that system.