We have come a long way now that we have these advanced classifiers. You would be surprised how low tech the initial product was, by low tech I mean devoid of any ML/AI. We went GA in end of 2019.
Saw a lot of interesting use cases too for e.g Japanese enterprises wanting to detect cases like suicide or intent to suicide, that is why we have multiple types of classifiers.
I worked on the Infra side (not ML). That too was “low-tech” or the more apt term would be “not the latest tech”. Core parts of the app were part of a monolith (think Exchange). Then we were using a really old .NET Framework version for our MVC app. Lot of the storage technologies we used were very MS specific as well. AFAIK, all of this is still valid today.