A few month ago Mario Fusco (Drools project lead) wrote “Combining LLM flexibility and rule engine predictability“: “The power and flexibility that a well-trained Deep Learning model gives you is virtually infinite, but often, at least in some parts of your application, what you need is more confined control to make it adhere to your business domain and rules. Given these premises why not mixing 2 very different and complementary AI branches like deep learning and symbolic reasoning? Moving forward with the mortgage example, this will give us a chance to implement an application with the corporate rigor required by the strict business rules of a bank, but queryable in the most human friendly possible way.” Link