Fraudex is in development — join the design-partner programRequest access →
Finmaro risk layerDeveloping

Fraudex

Realtime anomaly and fraud defense for your money.

Fraudex is designed to watch every connected financial flow—not only one bank or card. It scores transactions, identifies duplicates and policy violations, builds counterparty context, and routes suspicious payments to human review before settlement where possible.

Cross-account monitoringCounterparty graphPre-settlement review
Transaction risk streamIllustrative monitoring
Priority review

Vendor payment differs from established behavior.

New bank account · invoice amount 3.8× vendor median · domain changed 2 days ago

Risk92
10:42:18
Northstar Logistics · $48,200
ACH · new destination account
Hold
10:39:04
Cloudline Hosting · $8,410
Card · expected recurring amount
Clear
10:31:52
Brightside Media · $12,000
Duplicate invoice number detected
Review
10:18:26
Payroll batch · $214,600
Bank · matches approved payroll file
Clear

Illustrative risk stream. Streaming ML and graph models are planned.

The visibility problem

Fraud hides between disconnected systems.

Bank controls see the bank. Card platforms see their cards. Fraudex is designed to see the whole cash picture and the relationships between vendors, invoices, accounts, and approvals.

Stream

Score every connected money flow.

Monitor banks, cards, AP, and Stripe continuously instead of waiting for month-end reconciliation.

Context

Understand the counterparty, not only the payment.

Graph relationships expose changed accounts, shared identities, unusual clusters, and possible collusion.

Control

Route risk into the approval path.

High-risk transactions can be held for review and passed to Treasa’s policy gate before execution.

How Fraudex works

From transaction event to an explainable decision.

Stream

Ingest the event

Normalize transactions from connected banks, cards, AP, and revenue systems.

Score

Detect anomalies

Compare amount, timing, identity, policy, and historical behavior in real time.

Graph

Map relationships

Evaluate vendors, accounts, domains, invoices, and shared counterparties.

Gate

Clear or escalate

Release expected activity or route suspicious payments to human review.

Planned technology

Risk detection has to run before settlement.

Streaming inference and graph analysis need enough throughput to score activity inline, while the decision can still prevent loss.

Development disclosure: initial scope is duplicate and anomalous payment detection with a human review queue. Cross-customer graph learning and automated blocking are planned and subject to consent and payment-rail constraints.
StreamingMorpheus realtime anomaly pipelinePlanned
GraphRAPIDS cuGraph counterparty analyticsPlanned
ModelscuML / CUDA risk scoringPlanned
DeploymentNIM + NVIDIA AI EnterprisePlanned

See the suspicious payment before it settles.