Building Recommendation Systems in Python and JAX

Part V. The Future of Recs

I’m hungry for more. What are companies doing in Prod?

We’ve come so far, but there’s so much more! RecSys moves fast, and it’s worth knowing the concepts you’ll see at next year’s conferences. These ideas are already proven to some extent—none of them are pure science fiction—but they haven’t settled into their final forms yet.

Before we get into these few very modern ideas, it’s also worth noting all the topics we haven’t covered in this book. Our most heinous omissions are probably reinforcement learning techniques and ideas related to conformal methods. Both are deeply important aspects of recommendation systems, but both also require a significantly different background and treatment, and thus are ill-suited to fit into the structure here. Additionally, neither is well introduced to the JAX ecosystem, and thus they are a much harder lift to scaffold.

When your scale is big enough, and the previous chapters are no longer buttering your toast, the following chapters will show you how to level up. At the time of writing, all these methods are in production at Fortune 500 companies valued at more than $10 billion. Learn these concepts and then go build the next TikTok.