premmunot007-premmu/FedSecure-Cart-FL-for-Secure-E-Commerce-Recommendation ? reverse-engineered prompt

Build me a privacy focused federated movie recommendation app called FedSecure Cart.

I want it to train a recommendation model across multiple simulated users, instead of keeping all the data in one place. The core should be a neural collaborative filtering system with federated averaging, and it should protect updates with three layers of privacy, first gradient clipping and Gaussian noise for differential privacy, then secure aggregation with masking, and optionally homomorphic encryption for extra protection.

Please make it work on a MovieLens style dataset with non iid client splits, and include a simple way to compare privacy and recommendation quality. I also want a basic evaluation setup that checks how well it stands up to attacks like gradient inversion and membership inference, so I can see the tradeoff between accuracy, privacy, and overhead.

If you need to look up current docs or best practices for the privacy pieces, go ahead and do that.