March 2026 Summaries

Capgemini, in collaboration with Neo4j and Databricks, is advancing the operationalization of data and AI platforms by providing enterprises with a comprehensive framework that integrates data governance, architecture, and domain expertise. This collaboration aims to transform raw data into connected intelligence, crucial for sectors such as financial services, life sciences, and industrial markets, where explainability and traceability are paramount due to regulatory requirements. By utilizing Databricks for data management and AI and Neo4j's graph intelligence, Capgemini facilitates the development of knowledge foundations that support analytics, generative AI, and mission-critical operations. The approach emphasizes the importance of graph-based knowledge assets, such as digital twins and domain-specific ontologies, to enhance decision-making processes, improve resilience, and enable explainable AI outcomes. Capgemini's expertise in industry transformation allows organizations to swiftly transition from architectural design to execution, ensuring that AI tools are not only innovative but also scalable and trusted across the enterprise.

Context engineering is a critical component of agentic AI systems, ensuring that agents receive accurate and relevant information at the right moment to maintain consistent multi-step workflows. This process involves several key layers, such as orchestration, memory, and tool integration, to prevent failures like hallucinations and context rot. Orchestration frameworks like LangChain and LlamaIndex play a significant role in managing agent workflows, with LangChain offering a comprehensive ecosystem for tool integration and state management, while LlamaIndex excels in data-centric applications requiring complex retrieval strategies. Memory layers, including short-term and long-term memory, are essential for maintaining context continuity and storing domain knowledge, often represented through knowledge graphs like those provided by Neo4j. The Model Context Protocol (MCP) standardizes tool connections to enhance agent reliability, while techniques like compression and chunking optimize context efficiency by reducing token waste. Evaluation frameworks in systems like LangChain and LlamaIndex ensure agent performance remains reliable by measuring relevance, safety, and accuracy. Ultimately, context engineering enables agentic systems to make grounded, auditable, and smart decisions by preserving the structure of domain knowledge and systematically managing context across various tasks.

Healthcare organizations are increasingly recognizing that the true barrier to effective AI implementation lies not in the AI models themselves but in the fragmented data infrastructure beneath them. As AI systems in healthcare evolve from proof of concept to reliable, scalable solutions, knowledge graphs have emerged as critical tools for addressing this challenge. These graphs integrate diverse datasets into a connected, explainable framework, enabling healthcare entities to maintain up-to-date, accurate insights crucial for patient safety, regulatory compliance, and fraud detection. Unlike static large language models, knowledge graphs adapt to evolving medical knowledge by continuously updating with new guidelines and regulatory changes, thereby serving as a dynamic enterprise memory. They provide a semantic backbone that facilitates interoperability and consistent insights across systems, ensuring that AI can move from pilot projects to impactful, production-ready applications. Consequently, healthcare CIOs are advised to adopt targeted, domain-specific knowledge graphs as part of their data fabric to accelerate time to value and enhance governance, positioning their organizations for competitive advantage in an AI-driven landscape.

Neo4j Community Edition offers a robust platform for building graph-powered AI and intelligent applications, supporting features such as vector search for AI workflows, enhanced graph visualization through the Aura console, and centralized management with Fleet Manager. It enables users to store embeddings directly on nodes and execute queries that combine relationship data with semantic similarity, facilitating applications like recommendation engines and fraud detection. The Cypher query language has been updated to align with emerging Graph Query Language (GQL) standards, offering more expressive and scalable graph workflows. Users can deploy Neo4j Community Edition to the cloud with a single click via AWS, Azure, or Google Cloud marketplaces, ensuring a production-grade environment within their own cloud infrastructure. The platform also supports visual exploration of graph patterns, which is particularly useful for data model validation and stakeholder communication. As organizations grow, they can seamlessly migrate to managed services like AuraDB to leverage fully managed cloud solutions, reducing operational overhead while maintaining control over data models and queries.

The blog post by Adam Cowley critiques the frequent misuse of the OPTIONAL MATCH clause in Cypher queries, particularly by large language models (LLMs), which can lead to inefficiencies like row multiplication and a lack of clear output structure. He highlights that while LLMs can generate executable Cypher, they often produce suboptimal queries for complex data retrieval. Cowley suggests replacing OPTIONAL MATCH with more efficient Cypher patterns such as list comprehensions and COLLECT {} subqueries, which avoid unnecessary row multiplication and provide clearer, more performant code. He also emphasizes the importance of understanding when OPTIONAL MATCH is truly necessary, advocating for native graph patterns that express intentions more directly. For developers who wish to refine their skills, he recommends GraphAcademy's free courses on Cypher that cover essential building blocks, subqueries, and query optimization techniques.

Enterprises across various industries are increasingly integrating Neo4j's graph intelligence platform with the Databricks Data Intelligence Platform to enhance their data and AI capabilities. This partnership aims to address the need for understanding complex relationships within vast datasets, which are crucial for informed decision-making in areas like fraud detection, healthcare, and supply chain management. Neo4j adds a relational dimension to Databricks' data lakehouse, transforming raw data into knowledge graphs that facilitate real-time reasoning and more accurate AI outputs by incorporating explicit relationship paths and structural context. The combination of Databricks' scale and governance with Neo4j's ability to model interconnected data allows organizations to build trustworthy AI systems that are aligned with enterprise data controls. This integration supports explainability, audit requirements, and the evolution of enterprise knowledge as it enables continuous improvement of AI applications in a governed, secure environment.

Neo4j, in collaboration with Snowflake, has significantly advanced the field of identity resolution by processing large-scale consumer data with enhanced speed and precision. Audience Acuity, an identity services provider, utilized Neo4j's Graph Analytics within the Snowflake AI Data Cloud to achieve a breakthrough in resolving 3.8 billion fragmented data records from multiple sources into unified consumer profiles. This integration leverages graph technology to efficiently manage complex relationships and significantly reduce processing time from weeks to under 24 hours, while also decreasing costs by eliminating the need for separate infrastructure. This graph-native approach, employing algorithms like Weakly Connected Components (WCC), allows for the automatic clustering of related records into a single identity, even when data points are inconsistent. This advancement not only improves operational efficiency but also opens up possibilities for deeper insights into consumer behavior through advanced audience segmentation and lookalike modeling, transforming data into a valuable strategic asset.

Agentic AI architecture represents a significant shift from generative AI by focusing on enabling AI systems to act autonomously in real environments, emphasizing the design of systems that allow for planning, action, observation, and collaboration. This architecture is crucial as AI systems transition from merely generating responses to executing workflows, requiring robust system designs to ensure predictability and safety, particularly in multi-agent setups. Key components of agentic AI systems include agents with access to models, tools, and memory, orchestration for managing multiple agents, and guardrails to maintain system boundaries. Various architecture patterns, such as single-agent, multi-agent, parallel, competitive, sequential, router, network, and hierarchical, cater to different use cases, providing flexibility in handling tasks from simple to complex. Knowledge graphs enhance these systems by organizing data into entities and relationships, improving the reasoning and retrieval capabilities of AI agents. Ultimately, the implementation of agentic architecture aims to create production-ready systems that are dependable, auditable, and capable of complex task execution, with practical advice focusing on observability, tool constraints, and appropriate use of multi-agent patterns.

Graph intelligence is revolutionizing various scientific and industrial domains by effectively modeling and analyzing complex data relationships, thereby facilitating breakthroughs in fields such as drug discovery, agriculture, and biotechnology. By employing graph databases, institutions like Munich's Hauner Children’s Hospital and biotech firm Basecamp Research are overcoming limitations of traditional relational databases to analyze massive datasets efficiently. Hauner's use of graph databases has improved the diagnosis and treatment of rare pediatric diseases, while Basecamp's BaseGraph has significantly increased the known number of proteins and enhanced understanding of protein evolution. Similarly, Novartis leverages graph intelligence to streamline drug discovery by identifying connections between diseases, genes, and compounds. Beyond science, graph databases are also crucial in enterprise applications, such as fraud detection, supply chain management, and network security, by providing real-time insights and uncovering hidden patterns. The adaptability and analytical capabilities of graph databases are trusted by major organizations, including many Fortune 100 companies, to drive innovation and solve complex challenges.

This blog post offers insights into the latest developments and discussions in the world of graph databases, particularly focusing on Neo4j, a popular graph database platform. It covers a range of topics, including GraphRAG systems, AI memory solutions, and the practical deployment of AI agents in production environments. The post highlights significant events such as the NODES AI global conference and various workshops and webinars aimed at enhancing understanding of graph technology. Key figures and community members, like Vincent Koc, are featured for their contributions to the field, and several resources, including a free ebook on Essential GraphRAG, are provided to deepen readers' knowledge. The post also touches upon innovative approaches to addressing the limitations of large language models (LLMs) through neuro-symbolic AI and graph-based memory, emphasizing the transformative potential of graph technology in AI applications.

The text discusses the challenges and solutions for achieving truly personalized customer experiences, particularly focusing on the limitations of traditional Customer Data Platforms (CDPs) and the advantages of using knowledge graphs. It highlights that while many organizations collect vast amounts of data, they often struggle to connect and analyze this data in real-time to provide seamless customer experiences. Companies like Tchibo and Prospa have turned to customer graphs, a type of knowledge graph, to map and understand the relationships between customers, products, and interactions in real-time. This shift from collecting to connecting data enables organizations to deliver more responsive and personalized customer interactions, improve retention, and enhance the effectiveness of AI tools by providing a comprehensive view of customer behavior. The text emphasizes the importance of moving beyond siloed data and static models to adopt a connected data approach that can adapt to evolving customer needs and provide actionable insights quickly.

Dijkstra’s algorithm is a fundamental tool for finding the shortest paths in a graph, applicable in various contexts such as exploring an old mine for treasure or optimizing routes in supply chains, mapping, and telecommunications. The algorithm calculates the shortest path from a starting node to all other nodes, updating these paths iteratively by considering visited and unvisited nodes. An example illustrates its application: starting from Cave A, distances are calculated to neighboring caves, updating as closer paths are discovered, until the shortest path to the destination, Cave F, is found. This method is utilized in real-world systems to efficiently move goods, data, and people through complex networks, including logistics, fraud detection, and IT infrastructure. Dijkstra's algorithm showcases its utility by modeling systems as networks, as demonstrated in a hands-on example using Neo4j Graph Analytics to optimize routes in the NYC Subway system.

The concept of the "analytic continuum" emphasizes the seamless and uninterrupted flow of analytical thinking, where analysts continuously ask and explore questions without distractions, thus enhancing decision-making in government settings. This ideal state is often disrupted by traditional data systems that impose rigid workflows, leading to delays and fragmented insights. Knowledge graphs, however, revolutionize this process by embedding the meaning of relationships within data, allowing analysts to explore interconnected information without the limitations of conventional systems. This approach not only provides a comprehensive view of mission outcomes but also enhances the capabilities of AI by grounding responses in real, connected data, thereby reducing errors and improving decision accuracy. Government agencies are increasingly adopting graph technology to transform their operations, from identifying hidden networks and vulnerabilities to optimizing logistics, ultimately preserving the analytic continuum and accelerating mission outcomes.

Neo4j has developed a suite of connectors to integrate with major data platforms, addressing the challenges of bridging relational and graph data structures. These connectors, which have evolved from community and personal projects into fully supported products, include integrations with AWS Glue, Apache Kafka, Apache Spark, Confluent, Databricks, Google Cloud, Snowflake, and Microsoft Fabric. The connectors utilize Neo4j's drivers, notably the JDBC Driver, to translate SQL queries into Cypher, facilitating seamless data ingestion and transformation. The blog reflects on the development journey of these integrations, emphasizing their role in enabling zero-downtime migrations, graph-powered eventing systems, and enhanced data analytics. The Neo4j ecosystem is further enriched with tools like the Data Importer, which simplifies importing large datasets, and the Neo4j JDBC Driver, which aligns SQL concepts with graph data models. Additionally, the blog highlights community contributions, such as the Liquibase Neo4j plugin and Neo4j-Migrations, underscoring the collaborative effort in expanding Neo4j's capabilities within the data ecosystem.

Agentic workflows offer a dynamic approach to handling unpredictable processes by allowing AI systems to determine the next actions based on real-time context and feedback, rather than following a fixed sequence of steps as in traditional deterministic workflows. These workflows are particularly useful in scenarios like fraud detection, where the path to a solution is not predefined and requires adaptive decision-making. Agentic workflows involve a loop of planning, tool usage, reflection, and orchestration, with each step being informed by previous outcomes and current conditions. They differ from non-agentic workflows, which might use language models in a fixed pipeline, by actively using tools and iterating with feedback to achieve goals. By incorporating reusable design patterns such as planning, tool use, reflection, and orchestration, agentic workflows balance flexibility with control, offering reliability and adaptability in complex environments. Knowledge graphs enhance these systems by providing structured context and multi-hop reasoning capabilities, which improve retrieval precision and decision traceability.

Enterprises are increasingly integrating Neo4j Graph Intelligence into Microsoft Fabric to enhance AI capabilities and derive more meaningful insights from their data. While Microsoft Fabric's OneLake platform centralizes data, traditional table-based structures often lack the contextual relationships necessary for complex AI reasoning. By adding a graph layer, businesses can transform static data into interconnected intelligence, allowing for advanced analyses such as identifying customer behavior patterns, predicting churn, and optimizing supply chains. Neo4j's integration offers features like Text2Cypher, enabling natural-language queries, and supports graph algorithms for dynamic modeling, such as Digital Twins and fraud detection. This relationship-first architecture ensures data remains within the Microsoft ecosystem while providing enhanced insights that can be visualized in Power BI or used with Azure OpenAI models, fostering a shift from mere data strategies to comprehensive intelligence strategies by 2026.

This edition of "This Week in Neo4j" explores various developments and events in the realm of graph databases, highlighting the winning project of the Aura Agents Community Challenge, PagerDruid, which utilizes graph reasoning to diagnose microservice incidents. The discussion extends to long-term risks in graph persistence, with Jasper Blues warning about potential performance and data integrity issues arising from choices in storage and schema evolution. Jeremy Adams delves into graph algorithms used in match-style games, while Marco De Luca examines building a secure GraphRAG platform using confidential computing to protect sensitive AI workloads. The newsletter also previews upcoming events, including the global NODES AI event and various workshops, and features community member Alix de Cremoux, who will discuss integrating NLP and graph technologies in financial contexts.

This blog post by Christoffer Bergman, Director of Engineering at Neo4j, explores the integration of AI with Neo4j through new Cypher functions that enhance AI application development. It introduces a new Cypher procedure for structured text completion, highlighting the potential of AI beyond simple chat experiences by incorporating structured outputs for application functionality. Bergman uses an anecdote about rolling a die to demonstrate the feature's concept of "predictable randomness." The post further delves into a practical application: an AI-powered dinner planner and recipe generator. This uses structured outputs to generate recipes based on user preferences, dietary restrictions, and past menu history, and stores them in a graph format for better organization and recommendations. The structured data allows for the creation of shopping lists and even potential social applications like a dating app based on shared recipe preferences. The blog concludes with a cautionary note about verifying AI-generated recipes for safety, emphasizing the importance of reviewing AI outputs critically.

Two technical communities in Berlin and San Francisco recently explored the architecture of memory design for agentic AI systems, focusing on the role of context graphs. While the Berlin event, AI Memory and Founders Night, emphasized the importance of explicit connections between entities, the San Francisco Context Graph Meetup defined a context graph as a knowledge graph enriched with decision traces and procedural knowledge. Both gatherings concluded that context graphs are crucial for AI systems to reason effectively and maintain structured relationships, which vector searches alone cannot achieve. The discussions highlighted that context graphs serve not only as a tool for connecting structured domain knowledge to decision histories but also as a means to address the limitations of current AI memory systems that lack persistence and explicit reasoning. It was noted that the challenge of building context graphs is more about knowledge management than engineering, requiring a foundational layer of knowledge elicitation and formal encoding. The events underscored the convergence of ideas across global engineering communities, with context graphs becoming an essential component of AI memory architecture, advocating for these practices to become as standard as writing unit tests in software development.

Neo4j introduces GRAPH TYPE as a preview feature in version 2026.02, designed to simplify schema enforcement by allowing a holistic declaration of the data model instead of managing numerous isolated constraints. This new feature enables users to define nodes, multi-label implications, and relationship connections in a single, coherent structure, enhancing data integrity and maintainability while maintaining the flexibility of the graph database. GRAPH TYPE supports commands like SET, ADD, ALTER, and DROP to manage the lifecycle of the enforced data model, providing advanced validation capabilities that ensure strict adherence to specified data connections and label requirements. The preview is available across various editions of Neo4j, including Enterprise and Aura, and encourages user feedback to refine its functionality before general availability.

Agentic AI in the pharmaceutical and life sciences sectors leverages graph technology to transition from fragmented data and isolated language model experiments to scalable, explainable, and production-ready AI systems. This approach addresses challenges such as data heterogeneity, connectivity, and unstructured content by using knowledge graphs, which align with the inherently networked nature of biology. A core principle discussed is the separation of reasoning and action, where large language models (LLMs) handle intent and context, while deterministic tools execute actions, enhancing reliability and traceability. The use of schema-first knowledge graph construction ensures that AI systems are grounded, with clear domain schemas guiding data extraction and linking extracted facts to their sources. This architecture, already implemented by companies like Novartis, supports various stages of the drug lifecycle, including target identification and safety monitoring, by providing a structured context for intelligent reasoning and a memory layer for enterprise knowledge. The focus is on creating intelligent, explainable systems that are integrated across research and operational dashboards, facilitating deeper insights and better decision-making in regulated environments.