Real screens. Real graphs. Not mockups.

Two agent patterns running on Neo4j: a fraud investigation agent and a claims and CRM intelligence agent. Each answers in plain language and renders the connected entities as a live graph. Every screen below is the working software.

Fraud investigation agent  ·  Neo4j + React

Claims & CRM intelligence agent  ·  Neo4j

How the agents actually work, under the chat window: a prompt-driven query engine that turns a question into a validated, read-only query and narrates the result with every number traceable to a row.

And the full stack under each bot. Two agents, two frontends, one discipline: known intents short-circuit to deterministic queries, only novel questions reach the LLM (PII-sanitized on the API boundary), the user sees full results because the bot is gated to their own database access, and a semantic cache learns past questions to trim the prompt over time.

The blockchain is just another graph.

The same investigative pattern as a fraud ring, on a different graph. A multi-agent on-chain forensics pipeline: 10,000+ ledger events across six chains, scam-token filtering, FIFO cost-basis reconstruction, liquidations hash-verified twice against pool logs, and live liquidation risk read straight from the lending contract, not an indexer. Below is a sample report on synthetic data, so you can see the output.

on-chain forensics — sample report (synthetic data)

The same build discipline runs a second live system: a decade of NBA data, queried in plain English, served over the same kind of OAuth-secured MCP layer. Public data in, plain-English answers out.

Agents query production data under the controls a DBA lives under.

One HTTPS port. Every request is identity-checked, entitled per service, and read-only enforced at the backend before it ever reaches a database. This is the reference pattern I deploy, drawn generic.

Companies already pay for this capability. Most of it goes to a staffing markup, not to the engineer doing the work.

The reference build. 58.8 million records resolved into 44.5 million entities.

An independent lab build, run solo on my own time. The full entity-resolution and knowledge-graph governance stack Vertex delivers: 58.8 million records resolved into 44.5 million entities on Neo4j, with survivorship, ontology enforcement, validation, access policy and audit on top. Every number below was measured.

%%{init: {"theme":"base","themeVariables":{
"background":"#0d1826",
"primaryColor":"#16293f","primaryTextColor":"#eaf2fa","primaryBorderColor":"#3d6690",
"secondaryColor":"#16293f","tertiaryColor":"#16293f",
"lineColor":"#7fa8cf","textColor":"#eaf2fa",
"clusterBkg":"#101f31","clusterBorder":"#3d6690",
"edgeLabelBackground":"#0d1826",
"fontSize":"16px","fontFamily":"ui-monospace,SFMono-Regular,Consolas,monospace"
},"flowchart":{"useMaxWidth":false,"nodeSpacing":45,"rankSpacing":60,"padding":14}} }%%
flowchart TB
  subgraph OUT["OUTSIDE THE GRAPH"]
    SRC["Source systems
CRM · SupportDesk · Snowflake"]
    TTL["customer.ttl
the ontology, in git"]
    WH["Warehouse
bulk rows never enter"]
    AG["Orchestrator / agent
MCP · External Service · Apex"]
  end

  subgraph BATCH["RUNS ON A SCHEDULE"]
    SR["SourceRecord
immutable, never edited"]
    PII["Email · Phone · Address · GovId
shared value nodes, not properties"]
    TB["Ontology compiled in
23 classes · 13,500 entailments"]
    MAP["Mapping registry
versioned · 29 bindings"]
    SC["Score · tier · hard guards
hard identifier vetoes"]
    CE["CanonicalEntity
one person, N sources"]
    PE["PersistentEntity
durable id · held across reruns"]
    SV["Survivorship
ranked per-attribute authority"]
    DE["DecisionEvent
why every choice was made"]
  end

  subgraph ASK["RUNS WHEN SOMETHING ASKS"]
    RT["Semantic router
graph · vector · push-down"]
    CA["Cache with literal guard"]
    QP["Query planner
traverse or push down"]
    TI["Governed tool call
policy checked · refusals recorded"]
  end

  SRC --> SR
  TTL --> TB
  SR --> PII
  TB --> SR
  MAP --> TB
  PII --> SC
  SC --> CE
  CE --> PE
  CE --> SV
  SV --> DE
  PE --> DE
  AG --> TI
  TI --> RT
  RT --> CA
  RT --> QP
  QP --> WH
  RT --> CE
  TI --> DE

  style OUT fill:#101f31,stroke:#3d6690,color:#9fc4e5
  style BATCH fill:#101f31,stroke:#4a7fb5,color:#9fc4e5
  style ASK fill:#101f31,stroke:#3f9c85,color:#8fd4c0
  style SC fill:#1b3550,stroke:#6ea8e0,color:#eaf2fa
  style PE fill:#1b3550,stroke:#6ea8e0,color:#eaf2fa
  style RT fill:#14332c,stroke:#4fb89c,color:#eaf2fa
  style TI fill:#14332c,stroke:#4fb89c,color:#eaf2fa
  style QP fill:#14332c,stroke:#4fb89c,color:#eaf2fa
  style CA fill:#14332c,stroke:#4fb89c,color:#eaf2fa

A clear ladder. Fixed price at every step.