We have a Q&A database. The questions, answers are both trigram indexed and also have embeddings. All in postgres. We then use pgvector + trigram search, combine them by relevance scores.