For at least a decade, we have established best practices for handling data and making sure it is delivered when needed and is correct. But with the explosion of LLMs and Transformer models, not all of those best practices apply anymore. In this talk, you will learn which best practices still make sense in this new Machine Learning world, and what new strategies you need to adopt to make sure that your LLMs can train properly and return inference in a reliable, timely and *accurate* manner.