FinGraphRAG: Building a Portfolio Risk & Market Analysis GraphRAG
Coding Nexus (Medium publication)
Saakshi Gupta
// ABSTRACT
A comprehensive GraphRAG system for portfolio risk and market analysis, combining vector search with knowledge-graph reasoning to provide explainable financial insights. Argues that relational financial questions (sector exposure, supply-chain risk, event impact) need graph reasoning, not just vector retrieval.
Summary
A long-form technical article published in the Coding Nexus Medium publication. Covers the design and engineering of FinGraphRAG — a GraphRAG system for portfolio risk and market analysis.
Core argument
Vector RAG is great for "find me documents about X." It breaks down on questions like:
- Which of my holdings are most exposed to the same supply-chain disruption?
- How might the new EU AI Act affect my portfolio?
- What % of my portfolio is exposed to lithium pricing?
These are relational questions. They need a graph, not just cosine similarity. The article walks through how to combine both: a Neo4j knowledge graph of companies/sectors/holdings/events, a Qdrant vector index over filings and news, hybrid retrieval that uses graph traversal to narrow candidates before semantic ranking, and a reasoning layer that produces structured answers with citations.
What the article covers
- The KG schema (and why entity resolution is the actual project)
- Hybrid retrieval mechanics
- Worked example: sector-exposure calculation
- Lessons from building it