Uses LangChain, gpt-4 and gpt-3.5-turbo to write an initial prompt, then iteratively (automagically) perfect it by testing against a small dataset until high accuracy is achieved. - View it on GitHub