Semantic search just doesn't seem practical in my situation (regular inserts/updates to the index with auto-generated embeddings and no tolerance for downtime/excess latency). That would be a pretty common scenario for anyone in ecommerce so I'm keen to see what feedback you get from other users as more people try semantic search. Something still feels wrong to me with how Typesense is managing that read queue while generating embeddings. I don't see the same problems when importing to a collection without embeddings. I can't explain the excessively large latency and I wish my C++ knowledge was better!