A great open source Machine Learning server, with high degree of customization
can be found here
This framework is very well
documented and can be used as a black box to resolve problems like classification, ranking, logistic regression etc..
Using this kind of OS stack, the purpose of any proprietary solution is shifted towards the presentation of the data to the ML engine rather than developing horizontal AI code..
I am a firm believer that this is the way to go, since as the domain matures, there are many specialized frameworks that can solve the algorithmic part of the problem, leaving to the user the responsibility to collect, clean and normalize the data and simply send them to the proper service, which will come back with the optimal response.
Note that there are several more similar frameworks (for pricing though), who try accomplish the same task... See here for example:
http://aws.amazon.com/machine-learning/
https://cloud.google.com/prediction/docs