March 2024 Summaries

Microsoft Fabric has introduced native integration with MS Fabric and Azure OpenAI Service, improving data management, GenAI results, and reducing AI hallucinations. Additionally, two new courses are available on Vector Indexes and RAG with Knowledge Graphs, as well as articles on Graph Clustering and Building a Social Network. The company is also hosting a Neo4j Research panel where users can share their experiences with researchers and influence the future of Neo4j products. Upcoming events include a livestream on April 2, conferences at AWS Summit in Paris on April 3, and various webinars and GraphSummit Series events. Leann Chen, a community member, has recently shared exciting content on knowledge graphs and generative AI. The company is also offering courses on Introduction to Vector Indexes and Unstructured Data, Knowledge Graphs for RAG, Clustering Large Graphs With CLARANS, and Social Seed - Build Your Own Social Network. The tweet of the week highlights the use of Neo4j knowledge graphs in a RAG pipeline.

The new course on GraphAcademy, Introduction to Vector Indexes and Unstructured Data, teaches users how to understand unstructured data using Neo4j and vector indexes. Students will explore unstructured datasets, create embeddings and vector indexes, and use them to search and understand the data. Through hands-on tasks, they will learn to process unstructured data, use chunking strategies, and create relationships between the data. After completing the course, users will be able to build a graph database of their unstructured data and query it using vector indexes.

Graph ML and GenAI Research from Neo4j have published a practical guide to constructing and retrieving information from knowledge graphs in RAG applications using Neo4j and LangChain. Graph retrieval-augmented generation (GraphRAG) combines the strengths of graph databases with vector search methods, enhancing the depth and contextuality of retrieved information. The authors provide a step-by-step tutorial on how to create a knowledge graph using LLMs, set up a Neo4j instance, ingest data, construct and retrieve graphs, and implement a hybrid retrieval approach that combines vector and keyword indexes with graph retrieval. The implementation includes an unstructured data retriever, a graph retriever, and a final retriever that integrates the two components. The authors aim to make knowledge graph generation more accessible and easier to use for RAG applications.

Knowledge graphs are well-suited for handling complex, multi-part questions because they store data as a network of nodes and the relationship between them. This connected data structure allows RAG apps to navigate from one piece of information to another efficiently, accessing all related information. The technique of combining RAG with knowledge graphs is known as GraphRAG. Building a RAG app with a knowledge graph improves query efficiency, especially when you’re dealing with connected data, and you can dump any type of data (structured and unstructured) into the graph without having to re-design the schema.

Neo4j has collaborated with Microsoft to integrate its graph capabilities into the Microsoft Fabric analytics platform and Azure OpenAI Service. This partnership enables organizations to combine structured and unstructured data, discover hidden patterns, and deliver enterprise-grade GenAI applications. The integration allows for the transformation of unstructured data into knowledge graphs, enhancement of contextual understanding with GraphRAG, provision of long-term memory for LLMs with vector embedding integration, generation of graph-powered insights as part of Fabric, and deployment of Neo4j Graph Analytics as a native Fabric workload. This collaboration has the potential to transform enterprise analytics and AI, driving 80% of data and analytics innovations by 2025. By combining Neo4j's knowledge graphs and generative AI capabilities with Microsoft's Azure OpenAI Service, organizations can unlock new insights from their data and drive business outcomes. The partnership also enables seamless integration between Neo4j and Microsoft Fabric, allowing users to tap into Neo4j's graph data seamlessly through the Synapse Data Engineering module, execute Cypher queries and visualize results with Neo4j Browser, and transfer data between Neo4j and Fabric using Data Factory and JDBC/ODBC drivers.

Neo4j is a graph database that uses Cypher as its query language. Developer Relations Engineer at Neo4j is sharing their knowledge on how to find the data types of properties in Neo4j, which they found was not straightforward and required searching documentation, Slack messages, and trial and error. The engineer explains that there are many data types available in Neo4j, including standard simple data types such as String, Integer, Float, Boolean, and DateTime, and that most of them match up to types available in other technologies. They also explain that the `type()` and `valueType()` functions in Cypher do not return the desired data type information, but rather a string representation of a relationship type or an expression's value type, respectively. The engineer then introduces two APOC library functions, `apoc.meta.cypher.type()` and `apoc.meta.cypher.types()`, which provide exactly what they were looking for: the data types of properties in Neo4j. These functions can handle single properties, multiple properties, or even lists of nodes as input, making them a valuable tool for developers working with Neo4j.

The Neo4j JDBC driver version 6 has been released, offering improved performance, scalability, and compatibility with various platforms and tools in the Java ecosystem. The new driver is built on a dedicated version of Bolt, the Neo4j network protocol, and implements the JDBC spec proper. It provides features such as schema information extraction, pluggable SQL translation architecture, and good developer experience. The translator can handle complex queries, including multivalued inserts and joins, and supports configuration options for customization. The driver is tested against GraalVM in native image mode and is fully compatible with existing applications built on the common Java driver or Spring Data Neo4j. It should be used for new applications and integrations that require JDBC capabilities, especially those involving data movement platforms and ETL tools.

This week's edition of "This Week in Neo4j" covers various topics including data modelling using Google's Gemini Pro, a curated list of videos about the GenAI Stack, GraphQL and building an AI-powered search engine. The workshop series continues with geospatial data, and upcoming events include conferences, meetups, and webinars. A featured community member is Jennifer Abel, who shares her expertise on deploying Neo4j and NeoDash. Additionally, there's a tweet of the week from LangChainAI about their new video series focused on understanding the RAG landscape.

The Keymaker framework is a low-code recommendations framework that enhances Neo4j-based application development by offering innovative next-best-action recommendations. Its unique architecture separates the logic 'engine' from the user interface, ensuring flexibility in application design and functionality. The framework seamlessly integrates with Neo4j databases, allowing for flexible configuration of analytical query pipelines tailored to specific business needs. With Keymaker, each engine or query pipeline developed is automatically equipped with a GraphQL API endpoint, streamlining the development of user interfaces using frameworks like React. The framework provides score-based recommendations to users based on business logic defined within an engine, utilizing techniques such as collaborative and content-based filtering. Each engine consists of one or more phases, each with a specific purpose in the pipeline, and has access to all engine inputs as Cypher parameters. Keymaker offers benefits including scoring and explainability, dynamic flexibility, rapid development, enhanced readability, extracting visualization paths, optimized performance, and resources such as documentation and videos for setup and usage.

The Needle StarterKit 2.0 is a comprehensive library designed to boost the development of Neo4j front-end applications, offering a variety of templates and components tailored to specific use cases across different industries. The new version introduces four new templates, including Cybersecurity, Movie, E-commerce, and Foundation, each providing a unique set of features and functionality. Additionally, three new components have been added, including a chatbot interface, connection modal, and header, which can be easily integrated into existing applications. To get started with the Needle StarterKit 2.0, developers need to install Node v16 or higher and Neo4j v5.X or higher. The kit is open source and Apache 2.0, allowing users to access the source code, contribute, and explore the provided templates and components. A hosted app is also available for users to try out the starter kit in action without having to deploy it in their environment. With the release of version 2.0, developers can accelerate their Time to Value while focusing on their specific needs, with plans for future releases including more templates and components, as well as how-to tutorials, guides, and videos.

Cypher Workbench is a set of tools that helps developers conceptualize, model, and work with graphs and the Cypher query language. It was released under Neo4j Labs in 2020 as a tool to support data modeling and graph development. The tool provides a workflow for capturing domain knowledge, creating models, validating queries, and iterating through the development process. It includes tools such as Business Scenarios, Model, Cypher Suite, Cypher Builder, and Databases, which can be used together to support various roles in the graph development process. The tool is available under the Apache 2.0 license and comes with no SLAs or implied support, but offers best effort support through the Neo4j community.

This week in Neo4j brings a range of topics including Cypher debugging, JSON-based agents for LLMs, and an introduction to graphs. The GraphStuff.FM podcast discusses RAG databases with Johannes Jolkkonen, who shares his expertise on when to choose a graph database versus alternative vector or relational stores. Additionally, there is a workshop series, conferences, meetups, training series, featured nodes, speaker spotlights, Cypher optimization tips, and more. The community is also engaging through live streams, tweets, and comments, with topics ranging from lessons learned in real-world graph app development to introducing the basics of Cypher and graph databases.

The text discusses the use of Google's Gemini Pro AI model to transform an Entity-Relationship (ER) diagram into a Graph Model stored in Neo4j. The ER diagram is used as input for the Gemini Pro model, which extracts entities, relationships, and fields from the diagram. The extracted data is then transformed into assets of a property graph model stored in Neo4j. The process involves using multi-modal prompts to include text, images, and video in prompt requests, and generating responses that contain recognized details of entities, relationships, and their fields. The response can be used as input for Neo4j's query language, Cypher, to create nodes, relationships, and constraints. Additionally, the process can generate LOAD CSV statements for ingesting entity records and relationship records into Neo4j. The text highlights the benefits of using graph databases for handling complex relationships and hierarchies, and the potential applications of generative data transformation in various domains.

The text discusses advanced retrieval-augmented generation (RAG) strategies for building more robust and accurate RAG applications. These strategies are necessary due to the limitations of basic vector similarity search, which only compares semantic content without regard for other data aspects. The article introduces a step-back approach to prompting, where the model first asks and answers a general question about the fundamental principle or concept behind a query, and parent document retrievers, which use larger documents as context while indexing smaller chunks of those documents for better representation. Additionally, it discusses available strategies such as typical RAG, parent retriever, hypothetical questions, and summaries. The article also provides guidance on implementing these advanced RAG strategies using the neo4j-advanced-rag template with LangChain templates, which can be used to deploy retrieval-augmented generation applications in just a few minutes. By following the steps outlined in the article, developers can build more accurate and contextual RAG applications for better user experiences.

Using knowledge graphs can help build user trust in LLMs by handling structured and unstructured text within a single database, reducing the work required to give the model the information it needs. Knowledge graphs capture information about data points or entities in a domain or business and their relationships, allowing for efficient storage and retrieval of both structured and unstructured data. A microservices knowledge graph can store information about people, teams, microservices, and tasks, enabling features like vector similarity search and Cypher query generation to find relevant tasks by name and description, and aggregate data using various grouping keys. A GraphRAG application with LangChain can support DevOps teams by leveraging the power of knowledge graphs to provide accurate and up-to-date answers to user queries. The code for this example is available on GitHub, along with other resources and learning materials.

This week's news from the world of graph databases includes updates on GenAI, a user-friendly Python library for Neo4j, SpringAI and the Neo4j PHP Driver. The workshop series is set to begin next week with sessions covering various topics. Additionally, the O'Reilly Knowledge Graph Course offers a free 30-day trial to newsletter subscribers. Upcoming events include NODES 2023 recordings, live streams, conferences, meetups, and training series. Featured speakers include Jacob Marquez, who will share his experience mapping customer journeys using Neo4j, and Adam Cowley, who demonstrated how to use langchain.js for custom GenAI apps.

The text discusses the concept of Retrieval-Augmented Generation (RAG) in the context of generative AI models, specifically LangChain. RAG is a technique for enhancing the accuracy and reliability of LLM-generated responses by grounding the model on external sources of knowledge to supplement the LLM's internal representation of information. The author focuses on implementing a retrieval query in LangChain using Python, which supplements or grounds the LLM's answer. They use the SEC (Securities and Exchange Commission) filings from the EDGAR database as their data set. The author constructs a retrieval query that pulls connected data of similar nodes, including Form, Person, Company, Manager, and Industry nodes, and returns the text, score, and metadata variables. The query uses Cypher queries in LangChain to find the most similar nodes and passes them into the retrieval query to pull additional context. The author provides examples of Cypher retrieval queries for Neo4j and demonstrates how to construct a retrieval query using LangChain. They also discuss the importance of mapping extra values in the metadata dictionary field to return the correct properties.

This week in Neo4j brings news and updates on various fronts. GraphAcademy has new content, including a Knowledge Graph Course from O'Reilly with a free 30-day trial for newsletter subscribers. The community is also looking forward to upcoming events such as NODES 2023, Women in Tech Day, and GraphSummit Series. Featured speaker Sebastian Lobentanzer will present on creating open-source frameworks for knowledge graph creation and LLM interaction. New courses are available on the GraphAcademy platform, including a course on building a Neo4j-backed chatbot with TypeScript. The newsletter highlights various tools and resources for working with Neo4j, such as the Neo4j Graph Algorithms Table of Elements and a Gen AI-powered song finder in four lines of code. Additionally, there's a video featuring Jim Webber discussing the limitations of GenerativeAI, as well as a tweet from Ed Carron showcasing his CompaniesHouseInterface tool for analyzing business networks.