This Week in Neo4j: Knowledge Graph, Life Sciences, GraphRAG, NODES, Startups and more

FEATURED COMMUNITY MEMBER: Johannes Sommer

KNOWLEDGE GRAPH: How to Keep Your AI Agent’s Knowledge Graph Clean

Paul Iusztin’s follow-up to his agent memory architecture piece focuses on the part most tutorials skip: keeping the graph clean as it grows. He separates entity resolution from deduplication, explains why collapsing the two is what silently corrupts graphs and adds two safety nets: a human review queue modelled as pending SAME_AS edges in the graph itself and a nightly dream pass that re-runs deduplication over recently ingested nodes to catch concurrent-write collisions.

LIFE SCIENCES: GraphTalk pharma & life sciences

GraphTalk Pharma & Life Sciences returns on July 2 – again as a hybrid event so that you can join in person or online. The day follows the pharma product life cycle with practitioner talks from Bayer (GraphRAG for drug target discovery), KWS Group (explainable candidate gene identification), BASF (knowledge graphs for non-model plants), Insel Spital Bern (multilingual EHR mapping in Neo4j) and more.

GRAPHRAG: Build a Semantic Layer from GCP with Neocarta

Alex Gilmore walks through Neocarta, a Neo4j Labs Python library that builds a graph-backed semantic layer by pulling schema metadata from BigQuery and business glossary terms from GCP Knowledge Catalogue – two CLI commands, and you have a traversable graph of databases, tables, columns and business terms with vector and full-text indexes built in. The post then wires the result to an MCP server so that any agent can perform a hybrid semantic search across your data landscape, rather than dumping raw schemas into the context window.

NODES: Neo4j Showroom

Alani Connect lets you discover curated talks, groundbreaking research and visionary leaders and connects you to the ideas, people and solutions driving real change. The Neo4j Showroom by Athulya Anil, William Lyon & Ashok Vishwakarma has all the sessions from NODES AI for you to explore.

STARTUPS: Now available: Dedicated Neo4j field engineers empowering AI startups

Brian O’Keefe explains what he’s hearing from founders: credits are fine, but what they actually need is a real technical conversation about schema design, production scaling and getting a prototype running fast enough to convince sceptical co-founders.
We want to give startup founders the right tools and support to succeed when building with Neo4j. From now on, I will also showcase a few of our Startup Program Members – First up: QuantumSpace

QuantumSpace is using Neo4j AuraDB to combat art forgery by extracting up to 60,000 data points from a single painting. They map elements such as colour, brushstroke physics and historical provenance as a graph. The relationships between these data points help establish authenticity, with Neo4j serving as their knowledge layer to connect verifiable facts with inferred signals such as tonal similarity and geographic context. What began as a thesis experiment has now attracted interest from the fashion industry, automotive sector, and even a European space agency.

POST OF THE WEEK: Spring I/O

Thinking in Relationships: Practical Graph Database Modelling in Java with Neo4j by @paulienvanalst.bsky.social @ Spring I/O 2026

Video: youtu.be/ImdfQzopSkg

Slides: www.slideshare.net/slideshow/pr… #springio26

[image or embed]

— Spring I/O (@springio.net) 9. Juni 2026 um 18:40

Please share it if you like it!