This talk will cover how we created a system to generate health insurance appeals from denials. We'll start with a quick look at the different attempts at model fine-tuning (along with what eventually worked well for us), then look at the middleware, how we integrate in-house and external models, steps to protect privacy, and finally, some truly questionable frontend code (or truly great frontend code if we finish the rewrite in time—you never know). If you love the American healthcare system as it stands today, this is probably not the talk for you. For those ambivalent about the healthcare system, that's totally fine; you can still see what it takes to bring an ML system from drinks at a bar all the way through to production.