RAG with a Neo4j Knowledge Graph: How it Works and How to Set It Up

In this video, Philipp Brunenberg explains how RAG (retrieval augmented generation) with a Neo4j Knowledge Graph works and how we can set it up on your laptop. He’ll also give a brief introduction on how we can customize the code from the Gen-AI Stack.
We’ll cover:
What is RAG?
How does RAG with a Neo4j Knowledge Graph work?
What are the benefits of using RAG & Neo4j?
How can I set up RAG with Neo4j using a Llama2 model and the GenAI Stack?
Links:
GenAI Walkthrough: neo4j.com/developer-blog/gena...
Neo4j & GenAI: neo4j.com/generativeai/
Github Repo: github.com/docker/genai-stack
RAG Blogs: neo4j.com/developer-blog/tagg...
LLM Blogs: neo4j.com/developer-blog/tagg...
@philippbrunenberg www.philipp-brunenberg.de/
00:00 Introduction
00:29 What is RAG?
02:09 The Neo4j Vector Store
03:03 Augmenting LLM answers with RAG
04:50 Using the GenAI-Stack
06:37 GenAI-Stack Configuration
08:08 The compose file
09:47 Loading data into Neo4j
11:59 Using the chat frontend
15:30 Conclusion
#neo4j #graphdatabase #knowledgegraph #langchain #genai #rag #ollama

[@aaen9417]

Thank you for publishing this type of content. This is absolutely amazing!

[@AAL3087]

Great whirlwind tour to present a concrete understading of RAG with knowledge graphs in Neo4j. Thanks. Looking forward to exploring this channel more to learn more about RAG/KGs/GenAIStack etc.

[@neo4j]

Thank you very much!

[@NelsLindahl]

Loved this video and the pace!

[@neo4j]

Thank you very much!

[@deepbayes6808]

This was really useful., also the presentor did an amazing job. But it was too short. I wish to see a longer more tutorial like version of it too. Or to have links to relevant knowledge and tutorials.

[@neo4j]

Thank you very much! we have some more in-depth material on our blog. Check the links in the video description

[@roy.awderby]

Thank you for the deep explanation

[@neo4j]

Thank you!

[@kenchang3456]

Thanks for this robust example.

[@neo4j]

Thank you!

[@tecnopadre]

It would be really useful to do this video adding how to convert the unstructured data to KG and the use it on a RAG.
Step by step it's what makes people use Neo4j more. Thank you!

[@neo4j]

Thank you for the suggestion!

[@joshuacunningham7912]

Excellent video. Thank you.

[@neo4j]

Thank you!

[@RajendraKadam725]

Thanks for sharing the video @neo4j.
By the way, both responses provided in the video were inaccurate.
Why wasn't the correct answer generated? 🤔

[@neo4j]

sorry - can you share the timestamp where you think the answers were wrong?

[@jacobschuster5261]

Brilliant, thank you!

[@neo4j]

Thanks! Glad you like it

[@allminhuman2602]

Great video! What would be the minimum recommended system specs for this to work?

[@neo4j]

Thank you! Any relatively modern laptop should work to give it a go and do a test-run.

[@leonh-kd3cm]

It is still not clear whether at the final stage what input do you provide to LLM to get the final question:
- Original Question text
- Text extracted from the KG nodes/edges
- SO, you do not provide text associated with the best K answers resulted form initial generation of embeddings from the user prompt?
...and how to you compose the text extracted from the nodes? What is the technique?

[@MatanItah]

About how long can we expect the StackOverflow loader to import using the settings in the video to run?

[@henkhbit5748]

Great video, I have Windows and ollama is not for windows yet. Can I run the docker under wsl2?

[@nbbhaskar3294]

I have wsl2 docker setup in my windows and I am able to run ollama successfully.

[@alimahmoudmansour9681]

Thank you so much

[@neo4j]

Glad it is helpful!

[@D1zZit]

Amazing! Planning on integrating this with my neo4j db which has 1 billion relationships, might be pushing the limits a bit 🤣🤣

[@neo4j]

let us know how it goes :)

[@samreenkhan4994]

did it work ?

[@D1zZit]

@@samreenkhan4994 Querying and using RAG on the database itself worked. Generating embeddings with FAST-RP did not despite the powerful specs of hardware. Instead had to generate embeddings on a condensed data form with many missing properties. Still working it out.

[@armelhyacinthe6173]

Thank you ❤

[@neo4j]

You're welcome! Glad it is helpful

[@googleyoutubechannel8554]

How does this compare to what MS is doing in the GraphRAG paper, it seems like MS was really jazzed on using the LLM, GPT4, (in some way it wasn't clear) to create the relationships, the edges/weights etc, and this was the secret sauce, I assume this is different from what Neo4j is doing, but could you expand?

[@neo4j]

this was a bit before GraphRAG become a term - check out the latest episodes from GoingMeta ( neo4j.com/video/going-meta-a-series-on-graphs-semantics-and-knowledge/ ) where go into GraphRAG a bit more as well

[@podunkman2709]

Is there any loader ready to insert own data (csv or else - not pdf) instead of stackOverflow?

[@neo4j]

not as such, but the repo is available and you can modify the loader to take on other data: github.com/docker/genai-stack

[@podunkman2709]

This is only demo of some commercial product?
Can I use it within company to build some service?

[@neo4j]

the stack is based on open-source components which you can use to build your own product/servcice

[@Aristocle]

How can I load PDF, markdown, into Neo4j?

[@neo4j]

not directly, see this community post for a few pointers: community.neo4j.com/t/loading-pdf-or-doc-content-into-neo4j/8967
this doesnt cover AI assisted tools which you can use these days for similar results

[@davidecacozza392]

Hi! It is possible to use an own neo4j graph?

[@neo4j]

yeah, totally - check out our recent Going Meta Episodes ( neo4j.com/video/going-meta-a-series-on-graphs-semantics-and-knowledge/ ) or our Courses on GraphAcademy ( graphacademy.neo4j.com/ )

[@samreenkhan4994]

did u try? if yes, then how

[@Sri_Harsha_Electronics_Guthik]

what is up with the page...? looks weird!