Neo4j Document Intelligence is a document intelligence feature based on Aura, released by Neo4j in June 2026. It converts PDF, DOCX, TXT, EPUB, HTML, and Markdown documents into searchable and queryable knowledge graphs. Researchers can provide documents and describe their desired analysis objectives in natural language, and the system will propose a graph schema and extract entities and relationships using large language models. If a document repository is simply a bookshelf for storing files, this tool is more like an exploratory map that connects concepts and evidence within the documents. The results can be used in a GraphRAG configuration, which utilizes both a lexical graph for original text search and an entity graph to represent semantic relationships. In traditional document-based knowledge graph construction, file parsing, chunk splitting, named entity recognition, relationship extraction, schema design, and graph loading all had to be implemented separately. In particular, research fields differ in their entity types and relationships, which led to problems where the initial graph model did not match actual questions, or where multi-step relationships scattered throughout the documents could not be explained by vector search alone. The key difference of Neo4j Document Intelligence is that it creates schema candidates from natural language instructions and constructs the relationship structure of the document within Aura, without requiring separate extraction code or prior graph modeling. Just as GPT generates answers based on the meaning of the text, this tool transforms the entities and relationships hidden in the text into nodes and links in the graph, creating a foundation for subsequent exploration. A life science researcher can input research papers in PDF format and Markdown research notes, and then request a schema in natural language that includes entities such as Gene, Protein, Disease, and Drug, and relationships such as INTERACTSWITH, ASSOCIATEDWITH, and TARGETS. In the generated entity graph, it is possible to trace multi-step paths to see how a specific drug is connected to proteins and diseases in multiple papers, and in the lexical graph, it is possible to search for relevant sections of the original text to confirm the evidence for the relationship. This is useful for organizing connections between documents that might be missed by simple keyword searches into candidate hypotheses or relationships to be reviewed. In addition, clinical and regulatory documents can be processed together, including DOCX protocols, PDF reports, and HTML guidelines, to structure the relationships between trial stages, evaluation variables, adverse events, and target diseases. The GraphRAG application can be configured to narrow down relevant paths in the entity graph and then search for the original text context in the lexical graph. However, since LLM-based extraction results do not automatically guarantee the facts in the original text, it is necessary to verify the sources for each relationship and have expert review before using them in life science and clinical decision-making. Supported limits, extraction model selection, data retention, and security conditions should be further verified in the official Aura documentation.

Neo4j Document Intelligence is a document intelligence feature based on Aura, released by Neo4j in June 2026. It converts PDF, DOCX, TXT, EPUB, HTML, and Markdown documents into searchable and queryable knowledge graphs. Researchers can provide documents and describe their desired analysis objectives in natural language, and the system will suggest a graph schema and use large language models to extract entities and relationships. If a document repository is simply a bookshelf for storing files, this tool is more like an interactive map that connects the concepts and evidence within the documents. The result allows for full-text search.

🔬 Literature-Based Drug-Target Knowledge Graph: Input research paper PDFs and Markdown research notes, and request Drug, Protein, Disease entities and TARGETS, ASSOCIATED_WITH relationships. Query the multi-hop paths of the entity graph and the original text segments of the lexical graph together to identify candidate drugs for repurposing.