Have a VM where I serve products. I want to rerank them based on a custom personalized transformer (personalized to user history). I understand this will add more latency if my VM server is on OVH and I have to query a runpod api endpoint to do inference. Any thoughts on my architecture?