Neo4j Inc. and Microsoft Corp. announced a collaboration today to integrate graph database features into Microsoft’s Fabric and Azure OpenAI services, with an aim to help users uncover patterns and relationships in structured and unstructured data that aren’t as easily visible in other data stores.

Azure OpenAI Service provides access to managed versions of various OpenAI language models. Fabric integrates existing Microsoft platforms such as Data Factory, Synapse and Power BI into a single software-as-a-service product.

Neo4j said the collaboration will enable Azure users to structure unstructured data and load it into a knowledge graph. From there, they can use Neo4j tools such as Bloom data visualization or the PowerBI connector for business intelligence.

For example, developers can use Azure Data Factory to ingest data from Microsoft’s OneLake data lake into Neo4j, extract data from the Azure Synapse Data Warehouse using the Neo4j data warehouse connector, run graph data science algorithms from Synapse data science notebooks and build interactive dashboards in PowerBI using Neo4j Knowledge Graphs.

New take on RAG

Azure users will also get access to GraphRAG, which uses a large language model to create a knowledge graph based on a private dataset. GraphRAG is an enhanced form of retrieval-augmented generation. It’s superior to conventional RAG for questions that traverse disparate pieces of information through their shared attributes and require a holistic understanding of summarized semantic concepts over large data collections.

An example of a typical use case for GraphRAG is a search that combines an unstructured document such as an automobile engine manual with a structured bill of material of the component parts. “You can do a dual search between a vectorized document search plus a graph search and blend it into one single answer,” said Sudhir Hasbe, Neo4j’s chief product officer.

Neo4j provides long-term memory for LLMs supporting native vector embeddings, which are numerical representations of typically non-numerical data objects that capture their inherent properties and relationships in a condensed format. They are widely used in natural language processing and recommendation systems. The software also supports vector storage and search capability. Developers can use Azure OpenAI embedding application program interfaces to create embeddings and store them in the Neo4j database.

“I can take a large document, chunk it using various technologies, use Azure OpenAI to create vector embeddings, and Neo4j can store it at every node or relationship level,” Hasbe said. He showed an example that parsed the Wikipedia entry for Microsoft Chief Executive Satya Nadella and displayed relationships between elements in the biography as graph nodes.

“We provide similar capabilities to any vector database now, with vector store and vector search built into the database,” he said. “We see more and more production workloads blending vectors with knowledge graphs in a single solution.”

Neo4j and Microsoft Fabric are collaborating on a project to deliver the graph database as a native workload for Graph Analytics on Microsoft Fabric. Hasbe said the companies hope to deliver a preview of the new capabilities during the second quarter of 2024.

Neo4j Chief Marketing Officer Chandra Rangan appeared at SiliconANGLE’s and theCUBE’s Supercloud 6 event to talk about how knowledge graphs are shaping the future of AI and cloud computing:

Image: Neo4j