van Zoratti, VP of Product Management, Neo4j
Scoprite le ultime innovazioni di Neo4j che consentono un’intelligenza guidata dalle relazioni su scala. Scoprite le più recenti integrazioni nel cloud e i miglioramenti del prodotto che rendono Neo4j una scelta essenziale per gli sviluppatori che realizzano applicazioni con dati interconnessi e IA generativa.
GraphSummit Milan - Visione e roadmap del prodotto Neo4j
1. Neo4j Inc. All rights reserved 2024
Visione e roadmap del
prodotto Neo4j
Ivan Zoratti
VP of Product Management
2. Neo4j Inc. All rights reserved 2024
SAFE HARBOR ROADMAP
DISCLAIMER
The information presented here is Neo4j, Inc. confidential and does not
constitute, and should not be construed as, a promise or commitment by
Neo4j to develop, market or deliver any particular product, feature or
function.
Neo4j reserves the right to change its product plans or roadmap at any
time, without obligation to notify any person of such changes.
The timing and content of Neo4j’s future product releases could differ
materially from the expectations discussed herein.
2
4. Neo4j
Product capabilities launched in 2023/2024
Neo4j Inc. All rights reserved 2024
5
● Parallel Runtime - faster analytical Queries
● Change Data Capture - better data integration
● Autonomous clustering & Fabric - limitless
scalability
● Graph Schema & constraints
● Backup with point-in-time recovery
● Incremental import
● Neo4j/AuraDB Ops Manager for managing
databases
● Aura Enterprise Database on all clouds
(AWS, GCP, Azure)
● SOC II Type 2 compliance, AuraDB APIs, RBAC
configuration
● Private Link & CMEK
● Log forwarding & performance metrics - better
observability
● Workspace - unified developer experience
● GraphQL Support & Simplified Drivers API
● Bloom support for GDS algorithms
● GDS Python API
● Knowledge Graph Embeddings
● Longest Path & Topological Sort Algorithm
● Vector Search & index
● Embedding APIs & LLM Models - Real Time
integration
● OpenAI + MS Azure OpenAI, VertexAI, AWS
Bedrock, Langchain, LlamaIndex etc. - Real Time
GenAI integration
5. Neo4j Inc. All rights reserved 2024
6
Cloud Scale
• Procure through Aura Console or via
Cloud Marketplace
• Zero maintenance, automated
upgrades and highly available
• Scalable and elastic, on-demand
• Enterprise-grade security
• SOC II Type 2 compliance
• Easier RBAC configuration with Aura
Console
• Private link
• CMEK
• Observability with Ops Manager,
performance metrics and logs
forwarding
6. Customer Managed Keys (Encryption)
7 Neo4j Inc. All rights reserved 2024
What is it
Aura encrypts all data at transit &
rest by default.
Customer Managed Keys (CMK)
is an alternative way to protect
cloud data for security conscious
Enterprises, enabling customers
to manage their own keys for
encryption / decryption at disk on
Aura using Key Management
Services (KMS) from their Cloud
Service Provider.
Why it is important
Customers can protect their own
data, control access and have
the ability to revoke access, even
from Neo4j.
Customers can adhere to their
own stringent security policy
around access and key rotation,
on top of Aura’s Enterprise grade
default security and compliance
posture.
9. Neo4j Inc. All rights reserved 2024
10
April 12, 2024
Welcome GQL!
GQL - Graph Query Language
The first new ISO language since 1987
GQL-fueled additions in Cypher:
• Node and relationship expressions WHERE
clause
• Richer label expressions
• Sophisticated pattern repetitions
• SQL-like synonims
• GQL Error codes
• GQL is Here: Your Cypher Queries in a GQL World
• GQL: The ISO Standard for Graphs Has Arrived
• ISO GQL: A Defining Moment in the History of
Database Innovation
10. Neo4j Inc. All rights reserved 2024
11
New constraints on nodes,
relationships and properties:
● Unique relationship
property
● Relationship key
● Property data types
NEO4J 5 NEW CAPABILITIES
Graph Schema
11. Graph Schema / Graph Type
Neo4j Inc. All rights reserved 2024
12
The definition of the informational content of a schema
(or rather a graph type), comprising:
● A set of node type descriptors
(also known as a node type set).
● A set of edge type descriptors
(also known as an edge type set).
● A node type name dictionary that maps node type
names,
which are identifiers, to node types contained in the
node type set of this graph type descriptor such that
each node type name is mapped to a single node type.
● An edge type name dictionary that maps edge
type names,
which are identifiers, to edge types contained in the
edge type set of this graph type descriptor such that
each edge type name is mapped to a single edge type.
CREATE OR REPLACE GRAPH TYPE FraudDet
(a:AccountHolder { FirstName :: STRING!,
LastName :: STRING!,
UniqueId :: STRING! }
...) REQUIRE UniqueId IS KEY,
(c:CreditCard {AccountNumber :: STRING!,
Balance :: FLOAT!,
...} ...) REQUIRE AccountNumber IS KEY, ...
(a)-[:HAS_CARD ...]->(c),
(a)-[:HAS_ACCOUNT ...]->(b),...
CREATE OR REPLACE DATABASE foo
...
[WITH GRAPH TYPE FrautDet]
...
12. Neo4j Inc. All rights reserved 2024
13
Graph Pattern
Matching
Improved expressivity of
graph navigation with
Quantified Path
Patterns,
a more powerful and
performant syntax to
navigate and traverse
your graph.
13. NEO4J 5.0 NEW CAPABILITIES
Database Enhancements
Graph Pattern Matching Example → Fraud Rings
Neo4j Inc. All rights reserved 2024
14
QPP
MATCH path=(a:Account)-[:PERFORMS]->(first_tx)
((tx_i)-[:BENEFITS_TO]->(a_i)-[:PERFORMS]->(tx_j)
WHERE tx_i.date < tx_j.date
AND 0.80 <= tx_i.amount / tx_j.amount <= 1.00
){3,6}
(last_tx)-[:BENEFITS_TO]->(a)
WHERE size(apoc.coll.toSet([a]+a_i)) = size([a]+a_i)
RETURN path
accountNumber:2
amount: 1000
date: 2023-01-01T10:10:10.000+0000
accountNumber:1
amount: 900
date: 2023-01-02T10:10:10.000+0000
amount: 729
date: 2023-01-04T10:10:10.000+0000
accountNumber:4
accountNumber:3
14. Neo4j Inc. All rights reserved 2024
15
Parallel
Runtime
Speed up
analytical
queries up to
100x
15. Neo4j Inc. All rights reserved 2024
16
Parallel
Runtime
Speed up
analytical
queries up to
100x
16. Neo4j Inc. All rights reserved 2024
17
Parallel
Runtime
Speed up
analytical
queries up to
100x MORE CORES
17. Neo4j Inc. All rights reserved 2024
18
Parallel
Runtime
Speed up
analytical
queries up to
100x
FASTER
QUERIES
MORE CORES
18. Neo4j Inc. All rights reserved 2024
19
BLOCK FORMAT
Memory Optimized
and Future Proof
An implementation of graph-native
that’s informed by more than a
decade of experience supporting real-
world production graph workloads.
Neo4j is still graph-first; block format
is:
• Native graph storage
• Optimized for connected data
• Index-free adjacency
Block format supersedes all previous
store formats.
Migrate, convert, import into Block
Format
20. Graph Data at Scale
21 Neo4j Inc. All rights reserved 2024
Autonomous Clustering
Easy, automated horizontal scale-
out
Composite Databases
Federated queries and sharded graphs
21. Graph Data at Scale
22 Neo4j Inc. All rights reserved 2024
22. Properties Sharding
23 Neo4j Inc. All rights reserved 2024
Users’ Connections TOPOLOGY DATABASE
SHARDED PROPERTY
DATABASES
Parallel
data load
Rolling
updates on
demand
23. AI Enabler
Graph Data Science & Generative AI
Neo4j Inc. All rights reserved 2024
24
24. Knowledge Graphs + LLMs
Facts
Explicit
Explainable
Words
Implicit
Opaque
KGs LLMs
+
Left Brain + Right Brain
Neo4j Inc. All rights reserved 2024
25
25. A Perfect Match
Artificial Intelligence
Machine Learning
Information Architecture
Data Architecture
LLM
Knowledge
Graph
Linguistic
Pattern
Matching
Hierarchical
Emergent
Features
Neo4j Inc. All rights reserved 2024
26
26. Could this be vector search?
Artificial Intelligence
Machine Learning
Information Architecture
Data Architecture
LLM
Knowledge
Graph
Linguistic
Pattern
Matching
Hierarchical
Emergent
Features
Neo4j Inc. All rights reserved 2024
27
27. Why RAG With Vector Databases Fall Short
Similarity is insufficient for rich enterprise reasoning
Neo4j Inc. All rights reserved 2024
28
1
3
2
4
Only leverage a fraction of
your data: Beyond simple
“metadata”, vector databases
alone fail to capture relationships
from structured data
Miss critical context: Struggle to
capture connections across
nuanced facts, making it
challenging to answer multi-step,
domain-specific, questions
Vector Similarity ≠ Relevance:
Vector search uses an incomplete
measure of similarity. Relying on it
solely can result in irrelevant and
duplicative results
Lack explainability:
The black-box nature of
vectors lacks transparency
and explainability
28. 29 Neo4j Inc. All rights reserved 2024
DATA INFORMATION KNOWLEDGE INSIGHT MEANING
records sets relationships patterns layers
What is a Knowledge Graph?
An information architecture with layered connections.
29. RAG with Neo4j
Neo4j Inc. All rights reserved 2024
30
Find similar documents,
content and data
Expanded context for
related information and
ranking results
Improve GenAI inferences and
insights. Discover new
relationships and entities
Unified search, knowledge graph and data science capabilities to
improve RAG quality and effectiveness
Vector Search,
Full-text Search,
Geospatial, Pattern
match
Data Science
Knowledge Graph
30. Knowledge Graph Complementary Benefits
LLM
Human
Application
Knowledge
Graph
Extend LLM
knowledge
through RAG
Invite human
exploration &
curation
Advanced
application
features & analysis
Neo4j Inc. All rights reserved 2024
31
31. 1 Knowledge Graph Construction
Gen AI use cases LLM
Knowledge
Graph
Neo4j Inc. All rights reserved 2024
32
32. Neo4j Inc. All rights reserved 2024
33
Knowledge Graph
Construction with
Cypher Templates
34. Human
1 Knowledge Graph Construction
Gen AI use cases LLM
Knowledge
Graph
2 RAG-based Chat Applications
Neo4j Inc. All rights reserved 2024
35
35. Neo4j Inc. All rights reserved 2024
36
Natural Language
Search combining
explicit and implicit
relationships
36. Neo4j Inc. All rights reserved 2024
37
Browser Co-Pilot
• Uses Text2Cypher
model provided by LLM
API service layer
• UI/UX improvements
underway for surfacing
the copilot feature in
Query
• Soon to be available in
Workspace / UPX
37. Application Human
1 Knowledge Graph Construction
Gen AI use cases LLM
Knowledge
Graph
2 RAG-based Chat Applications
3 RAG-enhanced General Applications
Neo4j Inc. All rights reserved 2024
38
38. Neo4j Inc. All rights reserved 2024
39
Natural Language
assistants and co-
pilots,
rooted in
business policy
Prompt +
Relevant
Information
Embedding API LLM API
User
Database
Search
Prompt Response
Relevant Results
Knowledge
Graph
Application
39. ● Integrate Neo4j with leading LLM
open-source frameworks such as
LangChain and LlamaIndex
● Call LLM APIs natively via Cypher
using our open-source APOC library
● Agnostic LLM orchestration
connecting graphs to OpenAI,
AWS Bedrock, GCP Vertex AI,
Azure, Anthropic, Hugging Face,
and other proprietary and open
source foundation models
Integrate with the GenAI Ecosystem
Neo4j Inc. All rights reserved 2024
40
GenAI Stack
Application
Generative AI & Embedding Models
Orchestration
Grounding Knowledge Graph
Neo4j GenAI Integrations
Text | Chat | Embedding
NL Query | Image Gen
Neo4j Drivers
Python JavaScript Java
Neo4jGraph
Neo4jVector
GraphCypherQAChain
Neo4jGraphStore
Neo4jVectorStore
KnowledgeGraphIndex
40. ● Co-Pilot in Neo4j Browser for autocomplete
& Cypher generation
● Bloom & NeoDash NL integration
● More framework integrations:
﹣ Langchain, LlamaIndex,
SemanticKernel, Spring.AI, Haystack
POWERING GENERATIVE AI APPS
Neo4j’s GenAI Roadmap
Neo4j Inc. All rights reserved 2024
41
Coming 2024+
41. Neo4j Inc. All rights reserved 2024
42
Grazie!
ivan@neo4j.com
Follow us!
@neo4j