I build an api that generates responses based on the responses received from GPT. It’s a simple api that builds a prompt based on the submitted parameters.
API + param’s → prompt building → GPT → json respons from GPT → building API response
I am testing various different approaches, wanting to find solutions that give the best results.
Unfortunately, it is difficult to assess the results of these changes. Currently, I generate and evaluate the results manually.
I would like to write a set of tests, specifying input → expected output. I think that with such a solution I could more easily assess the quality of the changes made.
What solution do you use?
Is there any tool / library to test the answers from GPT?