Unlocking AI's Hidden Connections With Graphs
Nyah Macklin · Neo4j · Day 2
Fetching from two API endpoints, loop and merge by id — but it breaks because the ids don't match.
The connection we wired by hand — and mismatched.
Now agents are making decisions about people. If the guess is wrong, we find out later in a warroom / post-mortem.
Jessica works at Apex Global — on the sanctions watchlist.
- Jessica requests a $25k credit line increase.
- The agent APPROVES!
Every fact was there, but the answer is still wrong. The connection was IMPLICIT, not EXPLICIT.
MIT — 95% of organizations report no measurable return.
- No place to keep the context.
- No proof of work.
- Miss the connection piece.
To overcome AI's black box problem, we need knowledge to be transparent.
Visual view of apple, vector view of apple, knowledge graph of an apple.
Knowledge graph view can be understood by human and AI.
- Find the connections.
- One hop? Or more than one hop?
Knowledge Graphs
Organized / visual representation of relationships between entities.
Property graph — nodes, relationships, properties.
Example Cypher query for a 4-hop compliance check vs. the SQL example.
Similarity
- Text similarity — documents with similar meaning.
- Structural similarity — finds entities with similar connections.
Almost nobody is building the second one.
Context Graph
Context graph — graph RAG.
Relatively new (Dec 2025).
Contextual difference — tracing the decision that was made.
Both context graph and knowledge graph find the decision path.
The missing "why" — no memory.
Traditional audit log vs. context graph.
- Records vs. why.
Where to Learn More?
- 1 — Build a system that cuts down on hallucinations.
- 2 — Force the agents to show their work — force them to walk a causal chain.
neo4j Graph Academy
graphacademy.neo4j.com
You: "I have no idea" -> full visibility