Kriv AI · Financial crime
Fraud Shield
Scores each card payment and wire as it happens with a trained model, adds velocity and policy rules on top, and shows the reasons behind every score: the signals that moved it, the counters it read, and what happens next.
Fraud Shield
Checking
Real-time fraud scoring
A trained Random Forest scores each transaction from 0 to 100, velocity and policy rules add points on top, and SHAP shows which signals moved the score.
The training set and model in the repository: 50 labelled transactions (17 fraud) from 17 countries, a 100-tree Random Forest over 10 features.
Tools
In this demo the model, SHAP and the velocity rules run in one container, with the velocity counters held in memory.
Event Hubs streaming, Service Bus alerts, the SQL audit log and Redis are built into the API and switch on when they are configured.
Transaction
Sample transactions
Pick a sample, change the amount or destination if you like, and score it.
Card-testing burst
Six $1.00 charges from one device, scored one at a time, so you can watch the counters arm. Then score the account-takeover wire on the Fraud Copilot tab and see what it inherits.
Loading…
Engine, model and velocity store
Checking…