← back
Why your agents need decision traces, not just documents — Zach Blumenfeld, Neo4j
Takeaway
Store and retrieve the reasoning behind past decisions so agents can apply precedents and explain recommendations.
Summary
- Context graphs supplement customer records and policies with previous decisions, precedents, and causal chains.
- A financial analyst demo combines Claude, OpenAI embeddings, Neo4j, and a Next.js frontend to retrieve evidence for a lending decision.
- Hybrid retrieval uses semantic similarity and graph embeddings to find structurally similar decision traces.
- The create-context-graph scaffolding tool generates an application with a selected framework, domain, and demo data; connectors can bring in external business records.
context-graphsneo4jdecision-traces
Original description
A knowledge base tells a financial analyst agent the risk factors. A context graph tells it whether to reject or accept, because it also carries past decision traces, the reasoning behind them, and how similar cases resolved. Zach from Neo4j walks through how context graphs extend a standard RAG setup with three layers: short term conversation history, long term extracted entities, and reasoning traces that embed into vectors so structurally similar past decisions surface alongside semantically similar ones. The fastest path in is `uvx create-context-graph`, a one-command scaffold that gives you a backend, frontend, demo data, and an MCP server. It ships with 22 built-in domains or generates a graph ontology from a custom domain you describe. The underlying `neo4j-agent-memory` package handles entity extraction through a spaCy to GLiNER to LLM pipeline with deduplication and merging baked in, and plugs into pydantic AI, LangGraph, Crew, Google ADK, and others. Speaker info: / zachblumenfeld