AFAIK it's fairly difficult to combine many machine learning models (although boosting is a thing). I'm a noob though.

We could do live online or batch training of models in a distributed manner if it is feasible. It would take more time to build new models because we would have to produce more data, but it's an idea.

For my personal use, free software has always been high enough quality; its at least 90% of the way there if not better. And I can trust it.