Hi Neo4J community
I am new so please take into account :-)
We are training LLM's with GraphRAG. We want to create ontologies and instances of our life science domain. Thease we want to define formally as a kind of "SSoT" or "Knowledge Graph" if you like.
These SSoT's are in part created by LLM training. And the SSoT's in turn direct and constrain the LLM response generation (as is standard practice).
As indicated in the title, we want to combine the strengths of both Graph representation and robustly defined and typed classes for objects. So we want to be able to define nodes as instances of such well defined classes. We mean in the same way that in Python we have instances of Pydantic classes or similar when using an alternative like JSON Schema.
We are aware that combining these capabilities into a single functional system is a frontier or research topic as manifested in this fine work on PG-Schema co-authored by @hannes_voigt
https://dl.acm.org/doi/pdf/10.1145/3589778
Here is my question:
What are our options suitable for adoption now to achieve a combination of these two capabilities. IOW, is there a well supported Neo4j based solution that combines Graph structure and class defintion for objects (to apply to the nodes). Of course we would require good LLM interoperbility.
Thank you for any suggestion or guidance. Happy to clarify the question if required.
My best
Eric