I am using GPT 3.5-Turbo-Instinct and feeding in my dataschema into the prompt. I am creating my SQL query using create_sql_chain and then running the query in the database through JDBC. So far, it is working okay however, if I am looking to push this to production, I need the model to not return anything if it is not confident in its answer or if the query being asked is too vague.

For instance, if I ask my text to sql model "how many rows are there", it should return back something saying "this query is too vague". Likewise, if I were to ask it a question such as, "Return the GPA and Networth of Nadir", how do I get it to do "WHERE LastName = 'Nadir' rather than "WHERE FirstName = 'Nadir'. I thought that by potentially getting confidence scores I could explore and find a threshold where the human needs to clarify what they meant until the model had a confident enough query