Ask HN: What are your tricks/infra working with embeddings?

I'm trying to design my infra for creating, storing, and retrieving embeddings in my AI applications and was wondering what are the different paths for it. I'm especially interested in NLP, but vision/multimodal could be interesting too.

Whether it's related to performance, scalability, or something else entirely, I'd love to hear your experiences and insights. Looking forward to your responses!