May I know what embedding model you're using? For eg, this demo uses sentence-bert and runs on 4vCPU and returns results in tens of milliseconds with 324K records: https://hn-comments-search.typesense.org