So we have an assistant plugged-in through the APIs, it works as expected. The only concern is that we want a mechanism where we can correct the output that it is generating.
In our use case, the assistant gets spawned and destroyed every cycle.
For eg. We send a query through the UI, it pulls a list of records from the database, and sends to the assistant, along with the instructions. 90% of the output is right. 10% is wrong, which is where we want to seek a way of correcting the output, and tell it where it went wrong. And do it until the output is exactly how we want it.
We can show the instructions, if needed.
Maybe the word is fine-tuning, but when the everything is integrated, how to fine tune?