Sadly, I have seen one. It was a vba script from the late 90s that used a simple dense multilayer network to do some unsupervised pattern classification. The linear algebra tools in vba/excel along with the solvers are all native dll code and the vba itself is all AOT compiled to native, so it typically runs very fast, and for small matrices it beats out numpy by an order of magnitude due to the ffi overhead. Was it the wrong tool? It depends on your constraints, but probably. It did work though.
Not strictly in the sense you mentioned (assuming that you mean "by themselves") but people may find [1] and [2] interesting.
[1] https://pub.towardsai.net/building-a-neural-network-with-bac...
[2] https://towardsdatascience.com/demystifying-feed-forward-and...