Create Knowledge Graphs from Unstructured Data with Neo4j Document Intelligence

Added: 2026-06-03

[00:00:00]Most organizations are sitting on a gold mine of information they can't actually use. It's in the documents, customer service logs, support tickets, call transcripts, internal memos, the connections between people, events, and decisions. And these documents are real.

[00:00:14]They're just invisible until something maps them. Introducing Neo4j Graph Intelligence. With Graph Intelligence, you ground the LLM of your choice with the knowledge graph built from your documents [music] and ask natural language questions get grounded, trusted results. But how does it work?

[00:00:31]In this example, we are playing the role of a trust and safety team at a leading financial institution. A compliance analyst noticed a circular transaction using Neo4j, >> [music] >> and we will use customer service calls from the affected accounts to dig a little deeper.

[00:00:47]Here's how you load data. Point Document Intelligence at your documents, PDFs, Markdowns, plain text. [music] In this demo, we're using call transcripts from the bank's wire desk.

[00:00:56]Then you give it an extraction prompt.

[00:00:59]Tell it what to pull out. Extract person, organization, account, create these relationships, normalize these names, keep these identifiers as strings. That's your schema. Document Intelligence does the rest.

[00:01:10]Your documents are now in a graph. We connected the Document Intelligence graph to an MCP server so we can investigate directly from a familiar LLM user surface. Let's ask about the three individuals tied to those accounts. From our call logs, you can spot the three-party loop behind circular transaction, and most importantly, who is behind it all. Ruben Dunn. He's no account at this bank. He doesn't appear in the know-your-customer system.

[00:01:34]>> [music] >> The rules engine never flagged him. But the call transcripts tell a different story. He's the authorized signatory on one of the cycle accounts. And the $700 that gets added to each leg of the cycle, that's his fee structured into the wire amounts [music] so it never shows up as a separate payment. That information was always in your documents. Document Intelligence makes it queryable.

[00:01:53]Find actionable [music] insights hidden in your unstructured data all on the world's leading graph intelligence platform, Neo4j.