Back in my day we called this real time training from implicit user feedback.

The engineering challenge here is an order of magnitude bigger though. An LLM is orders of magnitude bigger than a recommender system model. Kudos.