Work/VaultGuard
AICustom Software Development

VaultGuard

A streaming fraud-detection engine that flags anomalies in milliseconds across millions of transactions.

vaultguard.app
<50ms
scoring latency
−97%
false positives
M+
tx/day scored

The challenge

The incumbent rules engine cried wolf constantly — blocking real customers while missing novel fraud patterns.

What we built

A streaming pipeline that scores every transaction in real time against a model that learns continuously, with human-in-the-loop review for edge cases.

The outcome

Sub-50ms decisions, a 97% cut in false positives, and millions of transactions scored a day without breaking a sweat.

Tech stack

Technologies we used

Server-side
PythonFastAPIPyTorchRedis
Client-side
ReactTypeScriptRecharts
Integrations
KafkaPlaidPagerDuty
Infrastructure
KubernetesGPU nodesPrometheus
Next projectCartwheelPay

Have a similar challenge?

Tell us about your project — we reply within 24 hours, and you'll talk to the people who do the work.

Start a projectSee our work