To do this we learn embeddings — dense vector representations — of SQL queries using three independent approaches, then apply those vectors to both tasks.