Structure-agnostic Knowledge Graph Extracting LLM


I want to fine-tune (or train from scratch if I must) an LLM that takes data of any structure, generates a knowledge graph from that data, and allows a user to interact with the data via a chatbot.

There are knowledge graph datasets out there but these are datasets that take in a given data of a given format and convert it into a knowledge graph. In other words, these datasets are structure-specific.

I want to fine-tune a model that converts data of any structure to a knowledge graph.

How can I go about it?