Skip to main content
← All projects
Dafe Ofoh / Senior DevOps & AI Engineer01 / AI / decision support

GraphRAG Decision Trace

A decision-support prototype that shows the evidence and graph relationships behind an AI answer.

Product prototype
GraphRAG prototype showing a trading-limit question, connected entities in a context graph, and a list of related decisions
Screen from the public GraphRAG repository. The financial example uses demonstration data. View full-size image ↗
In practice

Trace a trading-limit exception

An analyst asks for precedent on a proposed trading-limit override. The screen brings the answer, its connected entities, and the recorded decision into one place.

  1. AskStart with a decision question in ordinary language.
  2. InspectReview linked graph context, policies, and similar precedents.
  3. TraceFollow the reasoning path and check the sources behind the answer.

The issue

An AI explanation is hard to trust when a reviewer cannot see which policies, precedents, and graph relationships informed it.

How it works

The prototype combines graph structure with semantic search. A Next.js interface calls a FastAPI service using Claude Agent SDK and Neo4j Graph Data Science, then presents a decision trace for review.

Discuss a project