Doesn't look like much. But under the hood it's a demo of an LLM-powered langchain agent using my Blawx software's public API to generate an answer based on a human-validated symbolic encoding, fully explainable with reference to source materials, and zero risk of hallucination. All of the usability upside of ChatGPT, none of the opacity and horseshit. It BARELY works with 3.5-turbo, some of the time. I need GPT4 API access to really generalize it, in addition to a few changes to Blawx to add natural language meta-data that can be used in prompts. If anyone has any tips on how to get access to GPT4 faster without breaking any laws, that'd be great.

I am creating and interpreter for AST stored in JSON (see picture) (Lisp-idea--style) and I am playing with partial & reactive evaluation (Excel--style) for it