:grey_question: How to RAG on Neo4J

[adriens]

:grey_question: Context

From Python/LlamaIndex and Python/Langchain I could successfully interact with a Neo4J instance :

• 🇺🇳 Discover a country UN SDGs concerns w/ Open Metadata on Neo4J
• With Lanchain
• With LlamaIndex

Now, I would like to use this on Quarkus for a corporate KG and I wonder how this could be achieved on the Quarkus stack.

:grey_question: Question

• Considering an input Neo4J configuration, show how to do RAG on it with Langchain4J and package the whole thing within Quarkus.

[geoand]

We don't yet have a Neo4j based embedding store, but it should not be hard to add one (especially since there is a Quarkiverse extension for Neo4j).

cc @jmartisk

[adriens]

We don't yet have a Neo4j based embedding store,

It would be partial embeddings, but also text2cypher as it should be able to use the schema to tranform the incoming question into a proper cypher query isn't it ?

[geoand]

That's a separate concern IMO

[adriens]

So here we should focus on the embedding, not the text2cypher : right ?

[geoand]

Indeed

[adriens]

Should I let this issue opened then ?

[geoand]

Yup

[jmartisk]

There's a neo4j embedding store in langchain4j upstream and it uses the same client as the quarkus-neo4j extension, so it should be possible to integrate here. Yeah I can look into it, unless somebody else is super eager to do this

[adriens]

That would be really awesome. I'll can provide you feedback while implemented as we are currently building a projetc on top of both QUarkus & Neo4J. :star_struck: :nerd_face:

[jmartisk]

Resolved via https://github.com/quarkiverse/quarkus-langchain4j/pull/548

[adriens]

Thanks a lot guys. I'll keep the community up-to-date. :information_source: We'll use this, it's still a work in progress, I'll embed this for a Call For Paper, which uses this extension

[geoand]

🥳

[adriens]

• https://github.com/quarkiverse/quarkus-langchain4j/pull/548