That's really interesting. What are the most important things you've learned to do with the LLMs to get better results? What do your problem descriptions look like? Are you going back and forth many times, or crafting an especially-high-quality initial prompt?
