On-Demand Webinar
From Documents to Networks: Graph Based RAG for Tax Investigation
Tax fraud rarely shows up in one document. A payment that looks routine on every invoice only becomes suspicious when you connect the companies, people and transactions behind it. In this session, Jacob Coles, a data scientist at Redfield, shows how GraphRAG built on local LLMs turns case files into a knowledge graph in Memgraph that investigators can question in plain language, without the data ever leaving their environment.
You’ll see a full investigation run live, from a case overview to the leads that still need checking, with an early look at Memgraph Lore.
What You’ll Learn
- Why basic RAG breaks on questions that aggregate records or follow connections, and how GraphRAG answers them
- How a local LLM extracts entities, amounts and relationships from memos and emails into a predefined schema
- How an agent writes and retries Cypher queries, shows its work and reports what it can’t verify
- Why small local models like GPT-OSS and Gemma hold up when the graph constrains the task
