PreferredAI/cornac ? reverse-engineered prompt

Build me a Python library for testing and comparing recommender system models, especially ones that can use extra data like item text, images, or social connections.

I want it to make it easy to load a dataset, split it for training and testing, run a bunch of recommendation models side by side, and print out common evaluation results like accuracy, ranking quality, and train and test time. It should include examples for classic matrix factorization style recommenders and be able to work smoothly with common machine learning tools if needed.

Also add a simple way to save a trained model and serve it through a small web API so I can ask for top recommendations for a user. If possible, include a few ready to run examples and clear docs so someone can try it quickly with a built in dataset. Look up current docs online if you need to.