What's a scalable way to handle data cleaning, while still keeping things simple?