The Seek AI workflow includes a human-in-the-loop component to ensure the accuracy of the output. However, having a human-in-the-loop for every single generation but puts the accountability on the analyst as opposed to the ML. This workflow ensures quality of results, helps train the model, and it increases efficiency for the data analyst by generating draft SQL. But once the model is trained, it becomes unnecessary to keep a human as part of the workflow.
Introducing confidence estimation and auto verify by Seek. With confidence estimation, Seek has an understanding of how accurate the SQL output is. If the confidence estimation is high enough, the platform will verify the insight and display the information immediately. Seek will indicate that this insight has been verified by Seek, not a human.
Seek provides configurable confidence estimate thresholds so customers can dial in the accuracy that is appropriate for their business. Anything below the threshold goes through the normal human-in-the-loop workflow to provide verification before returning insights to users.
