When do 'ask', the process here is to use vector database to get the most related DDL. But there is a situation that all of the ddl has nothing to do with the user question, which means that the final SQL Vana generates will definitely WRONG!
So I wonder if there is any mechanism or post-process to check or evaluate the retrieved ddl.🤔
What's more, when you do embedding, the raw ddl is not a natural language so the problem between ddl and user question is that they cannot be aligned in semantic space. It may make the result inaccurate!
When do 'ask', the process here is to use vector database to get the most related DDL. But there is a situation that all of the ddl has nothing to do with the user question, which means that the final SQL Vana generates will definitely WRONG!
So I wonder if there is any mechanism or post-process to check or evaluate the retrieved ddl.🤔
THX!