July 2025 Summaries

The Neo4j GraphRAG package for Python, now in version 1.8.0, introduces advanced schema-driven capabilities to facilitate the construction of knowledge graphs from unstructured data using large language models (LLMs). This package offers features such as automatic schema inference from text, flexible schema enforcement options, and granular control over the extraction process, making it accessible to both novices and experts in knowledge graph construction. The package is particularly useful for applications like cyber threat intelligence, where it can extract insights from advisories by creating knowledge graphs that highlight key entities and relationships. Users can customize their schema workflows by providing hand-crafted schemas or allowing automatic inference, with detailed logging available to track each step of the process. The package also includes built-in graph cleanup operations to ensure the quality and consistency of the final graph, providing transparency and control over schema enforcement at various levels.

The blog post outlines the process of building a GenAI chatbot capable of interpreting technical documents using Neo4j and Unstructured.io, focusing on the energy industry. It describes how Unstructured.io facilitates high-resolution chunking of documents, extracting tables and images, and how a Neo4j knowledge graph maintains the chunk sequence for context. The chatbot, built with Neo4j's Needle Starter Kit, utilizes the Neo4j GraphRAG Package to perform retrieval and generation tasks by implementing vector and full-text indexing on document chunks and entities. The system enhances GenAI accuracy by enabling semantic searches, full-text searches, and nearest-neighbor traversals through a lexical graph structure, which allows for detailed and inspectable responses. The post also highlights the use of OpenAI for embedding and entity extraction, and the use of metadata to filter out irrelevant images, thereby enhancing the chatbot's reliability and specificity. The approach, while demonstrated on energy documents, is adaptable to any technical document corpus with minimal code modification.

Noah Mayerhofer's article explores the process of converting unstructured text into knowledge graphs using large language models (LLMs), emphasizing their potential to automate the traditionally labor-intensive task of structuring raw data. Knowledge graphs, which organize information as interconnected networks, offer advantages over traditional databases by revealing relationships among entities like people, places, and events, thus enhancing the retrieval of insights from unstructured data. The article outlines a three-step process for this transformation: extracting nodes and relationships, performing entity disambiguation to merge duplicates, and importing data into Neo4j, a graph database. Despite challenges such as unpredictable LLM output, performance limitations, and transparency issues, the use of LLMs for knowledge graphs is shown to be viable and beneficial, enabling more intelligent applications like personalized recommendations. Mayerhofer suggests tools like the LLM Knowledge Graph Builder and Neo4j Data Importer to facilitate this process and highlights the evolving ecosystem of resources that support the integration of LLMs with knowledge graphs.

Neo4j Spatial, a plugin for the Neo4j graph database, facilitates the integration of geospatial analysis within a graph database environment, utilizing the Cypher query language. This exploration delves into the capabilities of layer management, spatial filtering, and intersection calculations, drawing parallels to traditional GIS tools like ArcGIS or QGIS. Users can create new layers for different feature types, perform spatial queries, and manage data efficiently through techniques such as using the R-tree index model for spatial indexing and employing variable-length path patterns for traversing data. The plugin also supports operations like point-on-polygon intersection, which involves identifying spatial relationships between features across different layers and updating the graph with new relationships, thus enabling the execution of GIS-type operations directly within the Neo4j environment. Additional functionalities include the use of labels to improve node identification and efficiency, as well as persisting derived spatial relationships back into the database, thereby enhancing the graph's utility for spatial analysis.

GraphTalk Pharma & Life Sciences 2025 focused on the application of graph technology, GraphRAG, and GenAI in the pharmaceutical and life sciences sectors, bringing together industry leaders to discuss solutions for existing data challenges. The event highlighted the use of Neo4j in various applications, such as Merck's Synaptix platform for managing R&D knowledge, Syngenta's NOCTIS toolkit for chemical reaction data, and LMU Klinikum's Project AMIGO for rare disease treatment. Bayer AG showcased its patient-centric data integration approach using graph models. The discussions emphasized the role of graph technology in enabling structured, scalable, and explainable solutions for complex biomedical data, enhancing capabilities in drug discovery, precision medicine, and more. The event underscored the growing importance of graph technology in AI-driven innovation within life sciences.

Matthew Wood, a Product Manager at Neo4j, explores the integration of geospatial analytics into graph databases in the first part of a blog series focusing on Neo4j Spatial. The series highlights the conceptual alignment between graph databases and geospatial analysis, emphasizing the benefits of combining spatial and non-spatial data to gain deeper insights. Neo4j Spatial, an open-source extension, enhances Neo4j's native capabilities by allowing the modeling and querying of spatial data like points, lines, and polygons. Wood details the process of installing the Neo4j Spatial plugin, creating and managing geospatial layers, and loading data into these layers using various formats such as shapefiles and CSV. He also discusses challenges encountered in handling different geometry types, like multipolygons, and the performance improvements achieved through concurrent transactions during data import. This exploration sets the stage for more advanced geospatial functionalities in subsequent blog entries.

Fraud costs the U.S. government significantly each year, with tax, claims, and contract fraud being particularly costly, and federal agencies like the IRS and CMS frequently targeted. As fraud schemes become more complex and global, employing tactics like Fraud-as-a-Service and synthetic identities, graph technology emerges as a powerful tool for prevention. Unlike relational databases, graph technology excels at modeling and querying intricate networks of relationships, allowing for faster detection of fraud patterns and connections. Federal agencies use graph databases to uncover hidden networks, streamline fraud detection, and enhance collaboration by creating comprehensive views of interconnected data, which are crucial for adapting to evolving fraud patterns. By integrating machine learning with graph technology, agencies can improve predictive insights, automate fraud risk prediction, and proactively identify suspicious activities, enhancing their overall fraud prevention strategies.

This week's edition of "This Week in Neo4j" highlights several developments in the graph database ecosystem, including the release of a new introductory video series on graph databases, upcoming live events featuring GraphRAG, and a spotlight on community member Alex Chantavy, who will discuss his journey from using Neo4j for internal tools at Lyft to founding a startup. The edition covers advancements in the Model Context Protocol (MCP) and its applications in AI-enhanced graph applications, offering a new course on GraphAcademy for developers. It also discusses the Text2Cypher task, which evaluates foundational language models' performance across multiple languages, and explores other educational resources like webinars and workshops aimed at enhancing users' skills in graph technology. Additionally, a post of the week humorously notes a unique observation about Pokémon types in relation to graph database setups.

Consumer credit fraud poses a significant challenge to the financial industry, costing over $10 billion annually, with traditional detection methods struggling to keep up with evolving fraud networks. To address this, a temporal graph modeling approach was implemented, particularly by a top-tier French bank, using a dynamic and scalable model that captures the state of fraud connections in real-time without future data leakage. This approach leverages a time-forest data model, which structures user interactions and shared resources in a way that preserves temporal dynamics, enabling accurate machine learning model training for fraud detection. By using Neo4j's Graph Data Science tools, the method optimizes the computation of weakly connected components, allowing for parallel processing and reducing the complexity of handling large graphs. This innovative model not only ensures the timely detection of fraud patterns but also offers a scalable solution suitable for production environments, demonstrating significant improvements in processing efficiency and accuracy.

A graph database organizes and stores data as a network of nodes and relationships, unlike relational databases that use tables. This structure allows for more meaningful analyses of connected data, making graph databases particularly effective for applications involving complex relationships, such as social media analysis, fraud detection, and supply chain management. For example, in a social media context, nodes can represent users while relationships capture interactions, allowing for a deeper understanding of user connections. Graph databases offer performance benefits by eliminating the need for JOIN operations, which can slow down relational databases as data volumes grow. They are flexible, allowing for easier schema modifications as data requirements evolve, and are especially advantageous in scenarios requiring the analysis of highly connected data. Despite these benefits, graph databases may not be ideal for applications with primarily transactional data, where relational databases might suffice. Neo4j is highlighted as a leading platform for graph databases, known for its native graph engine, scalability, and rich community support.

Christoffer Bergman, Director of Engineering at Neo4j, introduces a new match mode in Cypher, called REPEATABLE ELEMENTS, which allows revisiting relationships in graph queries, enhancing the traditional DIFFERENT RELATIONSHIPS mode that restricts such revisits. This capability is especially useful in scenarios like finding flight routes in a global airport dataset, where traversing the same connections multiple times can help solve complex queries. For instance, it allows finding multiple travel paths from Malmö Airport or overcoming the limited connectivity of Ísafjörður Airport in Iceland. The blog post demonstrates how REPEATABLE ELEMENTS can optimize travel itineraries for mileage runners or facilitate visits to various destinations with constraints on the number of flights. Bergman illustrates the use of this feature through detailed examples, explaining its potential to address challenges that were previously unsolvable due to the limitations of the traditional match mode.

Neo4j has been recognized as a 2025 Gartner Peer Insights Customers’ Choice in the Cloud Database Management Systems category, reflecting broad customer satisfaction and trust based on feedback rather than analyst-driven evaluations. This accolade highlights Neo4j's commitment to customer-centric innovation, with enhancements in performance, scalability, and enterprise readiness, such as the introduction of Parallel Runtime and efficient bulk operations. Their cloud offering, Neo4j AuraDB, has seen over 25,000 active databases, indicating strong market adoption and customer-driven development, including robust security features and cost-effective options. Additionally, Neo4j's GraphRAG innovations enhance GenAI applications by combining AI technologies with structured data frameworks, meeting enterprise demands for accuracy and explainability. The company’s collaborative efforts with major cloud platforms like AWS, Google Cloud, and Microsoft Azure ensure seamless integration, reinforcing Neo4j's role in solving complex data problems for over 1,700 customers, including major corporations like Klarna and Intuit. This recognition is seen not just as an acknowledgment of their technology but as a testament to their partnership with customers in addressing real-world challenges.

Knowledge graphs, built on graph databases, offer a solution to the "data model problem" by capturing relationships and business rules directly in the data structure, unlike traditional relational databases, which often bury these in SQL code. These graphs consist of nodes (data entities), relationships, and organizing principles that provide a flexible way to manage highly connected data. To build a knowledge graph, one must define its use case, choose a suitable database management system (either triple stores or property graph databases), model the data, prepare and ingest data into the graph, and then test and optimize the graph's performance. Real-world applications of knowledge graphs include enhancing recommendation engines, fraud detection, and enterprise search, as demonstrated by organizations like NASA and Cisco. These graphs evolve over time to accommodate new data and business needs, offering a scalable and efficient method for uncovering insights in complex datasets.

The text discusses the limitations of traditional relational databases in capturing the complex web of relationships inherent in business operations and highlights the advantages of using graph technology for data modeling. Graph databases, unlike relational databases, inherently store data in networks of interconnected relationships, allowing businesses to uncover hidden patterns, streamline operations, and gain deeper insights without the computational overhead of complex queries. The article outlines how graph technology can represent seven key domains within an enterprise, such as customer interactions, transactions, and employee networks, to provide a holistic view of organizational data. By integrating these domains, businesses can enhance decision-making, improve operational efficiency, and prepare for the AI-driven future, as evidenced by real-world examples from companies like Prospa, Boston Scientific, Dell, and Klarna. These companies have leveraged graph databases to transform their operations, demonstrating the potential of connected data to drive innovation and maintain a competitive edge in a rapidly evolving technological landscape.

The blog post discusses the evaluation of large language models (LLMs) on the Text2Cypher task, which involves converting natural language questions into Cypher queries for Neo4j graph databases, with a focus on multilingual performance across English, Spanish, and Turkish. The authors released a multilingual test set and analyzed model performance, finding that LLMs perform best in English, followed by Spanish and Turkish, due to variations in language resources and linguistic similarities. The study showed that translating prompts had minimal impact on performance, while schema elements remained in English, suggesting future research could explore fully localized setups and language-specific tuning to improve cross-lingual query generation. The findings aim to promote broader research in structured query generation and contribute to the multilingual capabilities of LLMs.

Mark "Woody" Woodhams has been appointed as the new Chief Revenue Officer at Neo4j, bringing over 30 years of experience in enterprise software sales from companies like NetSuite and Oracle. Known for his ability to scale high-performing global teams and focus on customer impact, Woody recently helped BlackLine triple its revenue from $200 million to $600 million. He is enthusiastic about the potential of Neo4j's graph technology, particularly in the context of the growing demand for AI solutions. Woody emphasizes a customer-first approach, advocating for cross-departmental collaboration and transparency in sales processes. He also values authenticity in business relationships, inspired by his own experiences with impostor syndrome. Outside of work, Woody is passionate about music and has been collecting vinyl records since 1976, frequently attending live shows.

In the Neo4j Developer Blog, Christoffer Bergman discusses the introduction of native support for conditional queries in Cypher 25, released in Neo4j 2025.06.0, which enhances the capability of the declarative query language by allowing more concise and understandable code for complex operations. Cypher, traditionally lacking imperative constructs like loops and conditions, now includes WHEN...THEN...ELSE logic wrapped in a CALL clause, significantly simplifying the restructuring of queries, such as reordering linked lists. This upgrade makes Cypher queries less verbose and more readable, though it doesn't introduce new computational capabilities, as everything achievable with the new conditional queries was technically possible before. Additionally, Cypher 25 introduces features like the removal of the mandatory WITH clause between write and read operations and the new FILTER keyword, which further streamlines query syntax.

This edition of "This Week in Neo4j" highlights significant developments and educational resources in the world of graph databases, focusing on the advancements in GraphRAG through Lettria's 20% accuracy improvement by integrating Qdrant's vector search with Neo4j's knowledge graphs. The update introduces the GraphRAG learning path on GraphAcademy, offering insights into building knowledge graphs and utilizing frameworks like LangChain. It also provides tools for Symfony developers to enhance Neo4j integration, alongside opportunities for community engagement through user research panels and upcoming events, including live streams, conferences, and meetups. Featured community member Seema Naswa is recognized for her expertise in AI-driven solutions, and her upcoming talk on building smarter knowledge graph pipelines for semiconductor chips is noted. Additionally, resources on Neo4j integration, continuous learning paths, and webinars are mentioned to help users deepen their understanding and application of graph technology.

Neo4j's latest update supports two Cypher versions: Cypher 5, which is frozen and will only receive bug fixes and performance enhancements, and Cypher 25, which is an evolving version that introduces new features with each server release. Cypher 5 offers stability, catering to users managing existing systems, while Cypher 25 targets developers seeking new functionalities. This bifurcation allows users to upgrade server versions without altering existing queries and facilitates a smooth transition between versions, as Cypher versions are now independent of server versions. The evolving Cypher 25 version introduces features like conditional updating and subqueries, ensuring that the language remains innovative and responsive to the needs of developers and users who require cutting-edge features, while Cypher 5 continues to be supported for stability and backward compatibility.