Need to create something like bigml for inhouse use in a mid-sized company

Company where I happen to work operate several niche social networks, social network games, etc. I was tasked to create simple prediction and clustering API for in house use. Where should I start? Is there a good library which I could just write a wrapper for starters? Sure I will add specific features later but for now I just need something that somehow works...

In terms of size so far the biggest dataset is a few millions of rows with tag-like features. Also we have a few netflix like users' votes matrices which are smaller in size (about 100k users and 10k items or something like that).

To clarify even more basically we have three tasks:

1. recommend items to users

2. recommend users to users

3. cluster users for further analysis.