Automation extraction of game / policy rules from natural language inputs using LLMs. Enables structural data mining of LLM's semantic understanding of the world into a more trustworthy, structured form.
Proof of concept: User inputs information about foods they like/ dislike/ enjoy together/ don't enjoy together, the LLM pairs these personalized facts with general knowledge about food items -> a knowledge graph of food preference and nutrition is generated for each user. Enables personalized recipe recommendations.