End-to-end demonstration of code generation and chatbot creation in the Blawx prototype I've been working on. The LLM generates a block-based, visual, domain-specific constraint logic program based on the text of the law and the ontology defined by the user, then you encode a question and an interview and you have a chatbot that doesn't have the black box problem of LLMs, but has the same easy interface. Very interested in anyone's comments.

In principle this seems to be very similar solution the black box problem of LLMs as my idea with Levlo: Put a human (non-developer) readable deterministic language between LLM and some sort of a runtime. So I certainly see the value! 🙂 I made a demo about my approach couple months ago as well: . The language in the demo is fairly "low-level" and designed for arithmethics and boolean algebra, but it could just as well be more of a logic programming language for law.