GraphRAG with Neo4j: A Quant’s Preview

On September 19, from 15:30 to 19:30 CEST, I'll be joining a hands-on, virtual workshop presented by Packt Publishing and led by Dr. Alessandro Negro, Chief Scientist at GraphAware — and I wanted to share why I'm looking forward to it.

The session is built around a simple provocation: plain vector search, the backbone of most "chat with your documents" tools today, is good at finding text that sounds similar to your question, but weak at answering questions that depend on how things are connected. The workshop's answer is to pair a knowledge graph with an LLM: build a graph of entities and relationships from real documents, let an agent translate natural-language questions into Cypher — Neo4j's graph query language — and let it reason across multiple hops rather than a single best-matching passage. The hands-on project is an explainable "financial advisor" style application — ingesting financial docs (filings, news, etc.), storing the resulting knowledge graph in Neo4j, and combining vector search, graph traversal, and agentic retrieval so that every answer can be traced back to a specific fact and source rather than just sounding plausible.

That combination — relational reasoning plus a visible trail back to the evidence — is exactly what I'm hoping to draw on. In my own work in quantitative finance, "the model said so" could never be an unquestionably acceptable answer, and I don't expect that to change as more analytical workflows get automated. What I want to walk away with is a concrete sense of where this kind of graph-grounded reasoning genuinely earns its place — ahead of or alongside the analytical tools I already rely on, supporting or extending what they do. I see its role in exploring various financial network-like structures and the qualitative opportunities and risks buried in various financial docs, rather than in producing pure numbers, a task that stays squarely in the domain of the math models built for it.

I'll eagerly share my impressions once the workshop wraps.

Registration link (40% discount, code OZI40, applied automatically): https://luma.com/z3mtzgrf?coupon=OZI40&utm_source=Ozimek

(Disclosure: Packt Publishing provided me a complimentary ticket for this event.)