April 2025 Summaries

GraphQL on Neo4j AuraDB simplifies modern application development by providing a seamless, efficient, and scalable approach to building applications that require flexible, real-time data access. It eliminates the need for separate servers and reduces operational complexity, improving performance. This integration enables precise, efficient data access, streamlined architecture, and seamless integration with the Neo4j graph model, making it ideal for applications with relationship-heavy queries such as recommendation systems, fraud detection, and customer 360. The technology has gained popularity among developers working with modern front-end frameworks like React, Angular, and Vue, and its growth is attributed to annual developer conferences, a strong open-source community, increasing adoption among enterprises, and Neo4j's commitment to GraphQL innovation.

The Text2Cypher task involves translating natural language questions into Cypher queries. Researchers explored different hard-example selection techniques, including complexity-based, length-based, and Cypher-specific approaches, to improve the performance of fine-tuned models in this task. The analysis showed that using a smaller, more targeted subset of data, prioritizing more complex or challenging instances, can significantly reduce training time and cost by more than half while improving efficiency without a drastic drop in performance. However, the highest Google BLEU and Exact-Match scores remain below the performance achieved with the full dataset. The convergence of fine-tuned models suggests that increasing data diversity and fine-tuning hyperparameters could further improve performance. Additionally, the behavior of evaluation methods highlights the need to analyze how different data subsets impact the model's ability to generate accurate Cypher queries during execution-based evaluation.

This week in the world of graph databases, Neo4j is at the forefront with its Model Context Protocol (MCP) gaining significant attention. MCP is a universal protocol designed to seamlessly connect AI models with external tools, data sources, and APIs, allowing for more interactive and context-aware AI systems. Additionally, there are various integrations and applications of Neo4j with other platforms such as Snowflake, GenAI, GraphRAG, and Databricks, showcasing its versatility in the field of graph databases. Various events and conferences are lined up, including NODES 2025, Livestreams, Meetups, and Webinars, providing opportunities for developers to learn and share their knowledge about Neo4j and related technologies. Furthermore, a featured community member, Yoan Sallami, is building tomorrow's LLM-based agent system using graphs and symbolic AI, highlighting the potential of graph technology in innovative applications.

A knowledge graph is a design pattern for storing, organizing, and querying entities and relationships, while a graph database is a type of database specifically designed to store data with relationships. A knowledge graph is built on top of a graph database, providing the structure and meaning to the connected data. Graph databases are ideal for managing connected data, offering advantages such as flexible data models, scalability, and efficient querying capabilities. Knowledge graphs use organizing principles to capture business rules or categories for the data, serving as a flexible conceptual structure that helps surface deeper insights. They are useful for intelligent applications like recommendation engines, fraud detection, and AI systems, providing a strong foundation for use cases involving connected data.

The text discusses the integration of large language models (LLMs) and graph databases to transform static risk assessment into a dynamic data-driven strategy in commercial credit risk assessment. The current process involves analyzing financial statements, leveraging ratios, and assessing industry and business model factors, which are then fed into internal risk rating models. However, much of this information is stored as unstructured data, introducing complexity and challenges such as bias in data interpretation and overlooking critical factors due to human error. Large language models can extract vast amounts of unstructured data and organize it into a database for faster analysis. Graph databases can be used to represent complex relationships between entities and are better suited than pure vector databases for this task. The text also introduces the concept of GraphRAG, a retriever that combines vector search and Cypher traversals to incorporate additional nodes and relationships, resulting in more accurate answers. This approach enables the creation of a smarter brain that thinks like an analyst, connecting the dots between seemingly unrelated disclosures, commodities reports, and environmental events to surface insights that traditional models might miss entirely.

The Neo4j LLM Knowledge Graph Builder back-end architecture and API overview provides a comprehensive framework for integrating large language models (LLMs) with graph databases, enabling the creation of meaningful knowledge graphs from unstructured data. The system is built on Python with FastAPI, leveraging LangChain's LLM Graph Transformer and various document loaders to process content from diverse inputs, extract entities and relationships, and generate graph documents stored in a Neo4j database. Vector embeddings are used for semantic analysis, ensuring efficient data retrieval and contextual understanding. This modular design makes the back end a versatile foundation for AI-driven conversational interfaces and advanced data interactions, with features such as chatbot systems, knowledge graph visualization, and graph enhancements. The API offers scalable, efficient solutions for loading, processing, and interacting with documents across different sources, with endpoints for uploading documents, extracting content, saving to the database, querying chatbot content, and retrieving processed data.

This article discusses building an intelligent movie search system using Neo4j and Google Vertex AI. By integrating vector embeddings into a graph database, it enables semantic search that understands movie descriptions beyond simple keyword matching. This approach enhances recommendation accuracy by capturing contextual and relational meaning within the dataset. The system uses a knowledge graph to store movie metadata, including titles, descriptions, genres, directors, and actors. It generates vector embeddings for movie plots using Vertex AI's text embedding models and stores them in Neo4j for efficient retrieval. The article provides a step-by-step guide on how to build this project, including loading data into Neo4j, generating vector embeddings, storing and querying embeddings in Neo4j, and running similarity search queries. By leveraging graph-based AI solutions like GraphRAG with vector search, the system achieves more context-aware and accurate movie suggestions, demonstrating its potential for unlocking deeper search insights and improving user experiences.

Neo4j, a graph database company, has several exciting updates and events coming up. The call for papers for the annual NODES conference is now open, with submissions accepted until June 15, 2025. This year's conference will feature sessions on various topics such as GraphRAG, Agents, Knowledge Graphs, and more. Additionally, Neo4j is exploring fine-tuned Graph Neural Networks and Large Language Models to improve the accuracy of domain-specific question answering through its GraphRAG system. The company is also launching a new tutorial on Boosting Q&A Accuracy with GraphRAG Using PyG and Graph Databases. Furthermore, Neo4j User Research panel invites users to share their experiences and influence the future of Neo4j products. The company will be attending several conferences, including JAX and Data Innovation Summit, and hosting meetups in various locations.

Google Cloud is teaming up with Neo4j to enhance knowledge graph technology, providing tools for building agentic applications that seamlessly integrate diverse database tools, including those powered by Neo4j. Two new Text2Cypher models, built on Google's Gemma 3 architecture, are available, enabling developers to unlock greater value from their knowledge graph data stored in Neo4j. These models can be deployed directly from Model Garden on Vertex AI and are also publicly available on Hugging Face. The collaboration aims to address the need for a trusted open-source Text2Cypher model to streamline natural language querying of graph databases, empowering developers to accelerate the development of graph-powered solutions. Google's MCP Toolbox for Databases has been enhanced with Neo4j knowledge graphs, allowing developers to build agentic applications that integrate diverse database tools and leveraging knowledge graphs in GraphRAG applications. Live demos will be available at Google Cloud Next 25.

Neo4j has partnered with Google Cloud to bring graph technology to air-gapped environments, featuring an air-gapped marketplace where Neo4j can be deployed without external connectivity. This collaboration enables enterprises to use advanced graph database workloads while maintaining complete security and isolation, addressing challenges in secure, air-gapped environments. By running Neo4j on Google Distributed Cloud, organizations can gain access to powerful graph analytics capabilities for fraud detection, risk assessment, and identity verification without exposing sensitive data to external networks. The deployment process is streamlined, taking just a few clicks, and customers can customize their configuration before deploying Neo4j Enterprise Edition. This partnership marks a significant step toward enabling secure, high-performance graph analytics in fully isolated environments.

Pattern matching is the practice of detecting specific structures, sequences, or connections within data. It involves searching for meaningful patterns — whether shapes, trends, or recurring relationships — within raw or semi-structured data. Pattern matching helps make sense of data by identifying recurring structures and uncovering insights by exploring unknown or emergent patterns. This technique requires knowing the patterns in advance and is typically exact, requiring a match to be either present or absent. It can be used to find answers in data that address business problems, such as optimizing supply chains, detecting fraud, personalizing product recommendations, and analyzing customer behavior. Pattern matching is particularly effective in graph databases, where it allows for real-time analytics directly on live operational data with minimal performance impact. Cypher, a declarative query language designed specifically for graph databases, makes it easier to express patterns that span multiple hops or involve optional paths, improving readability and performance compared to SQL. With pattern matching, you can quickly find the solution to complex business questions by defining and detecting structures directly in your data.

Neo4j is expanding its partnership with Google Cloud to provide a comprehensive graph platform for businesses. The integration enables organizations to unlock the power of connected data, AI-driven insights, and cloud-native architectures, driving faster innovation and decision-making. Neo4j's graph capabilities are now integrated across Google Cloud services such as Distributed Cloud, GenAI Toolbox, Firebase Genkit, Dataflow, and Terraform, making it easier for developers to build smarter applications without needing deep database expertise. This partnership aims to make graph technology more accessible and empowering businesses to harness connected data for better decision-making.

The impact of schema representation in the Text2Cypher task is a crucial aspect of natural language-to-Cypher query translation. The use of different schema formats can significantly influence performance, with complex schemas posing challenges for large language models (LLMs). Schema linking and filtering techniques have been explored to address these challenges, with various approaches including exact-match, similarity-based matching, and named entity recognition (NER) masking. Experimental results show that pruning schema lengths can lead to significant cost reductions and improved performance, with the best approach being Pruned By Exact-Match Schema. Further exploration of schema filtering methods is recommended for specific datasets or practical applications, and additional research is needed to determine the effectiveness of these approaches on a range of LLM models.

Neo4j offers a range of graph database options to match different stages and requirements, including the Community Edition with free, open-source capabilities for learning and development, Enterprise Edition for self-managed control over regulatory or compliance requirements, and AuraDB as a fully managed cloud service offering enterprise-grade security, compliance, and scalability. Teams should consider moving beyond Community Edition when their projects become production use, require higher reliability, need to scale beyond single instances, or require expert support. Successful upgrades have been made by Milanote and Flaminem, who achieved performance improvements and transformed user experiences with Neo4j Aura Pro and Enterprise Edition, respectively. By choosing the right Neo4j offering, organizations can unlock insights from connected data that traditional databases miss.

The Model Context Protocol (MCP) is a universal protocol developed by Anthropic to connect large language models with external data sources, tools, infrastructure, and APIs as context for user workflows. It enables seamless integration of AI applications across different platforms and systems, reducing the need for individual, one-to-one integrations. MCP has gained significant traction since its launch in November 2024, with tens of thousands of server implementations and adoption by major players like Neo4j, OpenAI, Google Cloud Platform, and Microsoft. The protocol offers a range of features, including tools, resources, prompts, sampling, pings, and notifications, allowing users to interact with AI agents and other systems in a more efficient and productive way. Despite some challenges, MCP is expected to become the de-facto standard for next-gen agent and tool interactions, driving efficiencies for users and vendors alike.