Recommendation Engine Architecture

I'm currently building a review application where users will be able to review certain items with 1 to 5 rating. The users can also have friends and followers from which they receive information in their news feed(what actions their friends have taken). I need to recommend some items based on collaborative filtering and content-based recommendations. I think that using MySQL database will have extremely bad performance because there are millions of items. That's why I think that setting up a recommendation engine is the way to go.

Having my first steps in machine learning here and spent the whole morning reading about it but still have some doubts how to structure the engine. From what I understood I need an engine to do processing and run the algorithms and can use a search database(in my case Elasticsearch) to index the results and access them from my app. I found prediction.io and neo4j.com as possible options to use as recommendation engine but can't figure out the pros and cons of both.

Looking for someone with more experience in the field to suggest which one will be better for this case.