Knowledge Graphs & GraphRAG - Essential Ingredients for Enterprise GenAI_Gartner Symposium Gold Coast, Australia_Dr. Jesús Barrasa

Knowledge graphs are emerging as useful and often necessary for bringing Enterprise GenAI projects from PoC into production. They make GenAI more dependable, transparent and secure across a wide variety of use cases. They are also helpful in GenAI application development: providing a human-navigable view of relevant knowledge that can be queried and visualised. This talk will share up-to-date learnings from the evolving field of knowledge graphs; why more & more organisations are using knowledge graphs to achieve GenAI successes; and practical definitions, tools, and tips for getting started.

1. Knowledge Graphs &
GraphRAG
Essential Ingredients for Enterprise GenAI
Dr. Jesús Barrasa | Head of Solutions Architecture, Neo4j

2. At the heart of every enterprise
challenge is an explosion of
data complexity
TODAY’S REALITY

3. Explore the connections in your
data to unlock deeper insights

4. Neo4j Inc. All rights reserved 20234
A ring…
or a money laundering scheme

5. Neo4j Inc. All rights reserved 20235
r a critical device in a networkor a critical device in a network
A star…

6. Neo4j Inc. All rights reserved 20236
or a community of people
A cloud…

7. A fundamentally new approach is needed

8. The Property Graph: Simply Powerful

9. Cypher (GQL): Pattern Based

10. ISO GQL ‘24: New ISO DB Language

ISO standardization of Graph Query Language
(GQL) ensures that property graph handling will
remain open as a de jure standard, similarly to SQL.”

—James Governor, Principal Analyst and Co-Founder, RedMonk

It marks a new chapter in the history of query
languages, revolutionizing the way we interact with
and harness the power of interconnected data.”

—Tony Holland, Communications Lead, ISO/ IEC Joint Technical Committee 1/AG 1

“
“
https://www.iso.org/standard/76120.html

11. Performance: Graph Native Storage

12. We found Neo4j to be literally
thousands of times faster than our prior
MySQL solution, with queries that
require 10-100 times less code

— Dr. Volker Pacher, Senior Staff Architect at eBay


“

13. Real-time fraud
detection
Supply chain
resilience
Customer
experience
Digital Twins
Drug
discovery

Digital twin

14. ?????? ??????
28.0 > 31.9 ???
??????

15. ●Hallucinations
●General reasoning vs Domain Reasoning
●Ethical and bias concerns
●Access to the most recent data
●Inability to verify answers or do attribution (!!)
●Lack of enterprise domain knowledge
●Privacy & security challenges
“Houston, we have a problem”

16. Why are LLMs hitting a barrier?
Probabilistic/
Nondeterministic
Black-box
LLM

17. GPT-3.5 marked the beginning of a
GenAI hockey stick moment
LLMs

18. LLMs
GenAI Go-to-production Barrier

19. LLM Answer
?
So how do we overcome the
limitations of LLMs?

20. How can we go
further?

21. LLM Answer
?
#1 Fine-Tuning
Better task-specific answers from existing knowledge
Base
LLM
Fine-tuned
LLM

22. Retriever LLM + Rag Relevant Answer
Relevant
Info
Query
External Database
?
#2 RAG
Retrieval-Augmented Generation

23. LLMs with
fine-tuning
Go-to-Production
use cases unlocked
with RAG
RAG
RAG works
It raises the ceiling, but it’s still there

24. So what’s Next?
Fine
tuning
RAG?

26. GraphRAG is RAG where the R path
includes a Knowledge Graph
What is GraphRAG?

27. Knowledge Graph:
What’s the Big Deal?

28. https://blog.google/products/search/introducing-knowledge-graph-things-not/

31. Neo4j Inc. All rights reserved 2024Neo4j Inc. All rights reserved 2023
Commerce Transactions Graph

32. Neo4j Inc. All rights reserved 2024Neo4j Inc. All rights reserved 2023
Customer Relationship Graph

33. Digital Twin Graph

34. Neo4j Inc. All rights reserved 2024Neo4j Inc. All rights reserved 2023
Product and Parts Graph

35. Why do knowledge
graphs help with GenAI

39. LLMs
RAG with Vector
GraphRAG unlocks AI
GraphRAG

40. https://www.gartner.com/en/articles/understand-and-exploit-gen-ai-with-gartner-s-new-impact-radar

41. https://www.sequoiacap.com/podcast/training-data-sebastian-siemiatkowski/
…we’ve started using extensively in house is Neo4j and graphs. And we’ve
also looked at how people collaborate on building great information
And then on top of that we have Kiki, who then explores that information
and brings it to life. So we can go and ask Kiki about anything!
And all of that is getting centralized into one
place and connected through the knowledge
graph. We’re seeing that that is having a
tremendous impact on productivity internally.
So Kiki is basically our own internal chatbot
based on that growing internal knowledge graph.
“
“
“
A fantastic success story

42. KG LLM +
Better together

43. The Benefits of
GraphRAG

44. #1: Answer Quality

Answers are more accurate, but
also more useful & complete

46. "Across all metrics, our method demonstrates
consistent improvements. Notably, it
surpasses the baseline by 77.6% in MRR
and by 0.32 in BLEU score, substantiating its
superior retrieval efficacy and
question-answering accuracy."

47. “GraphRAG shows substantial improvement…demonstrating intelligence or mastery that
outperforms other approaches…. GraphRAG enables us to answer important classes of
questions that we cannot attempt with baseline RAG alone.”

48. #2: Easier
Development*

49. Easier Development
Opaque & Implicit
Customer: "I actually already fixed a couple of bugs
thanks to this!"
Transparent & Explicit

50. #3: Explainability
and Governance

51. Explainability &
Governance
“Neo4j enables us to enforce robust information security by applying
access controls at the subgraph level.”
-Eric Costantini, Chief Business Officer at Data^2.

52. 1. Quality &
Accuracy
2. Easier
Development
3. Explainability
& Governance
The Benefits of GraphRAG

54. Want to Learn More?

55. https://www.deeplearning.ai/short-courses/knowledge-graphs-rag/

56. https://graphacademy.neo4j.com/

57. https://neo4j.com/knowledge-graphs-practitioners-guide/
Free e-book!

58. Thank you!
Meet us at booth #907
Dr. Jesús Barrasa | Head of Solutions Architecture, Neo4j