Financial crime / Applied machine learning / Streaming systems

Real-Time Fraud Detection

A live investigation workspace. Stream synthetic payments, inspect calculated risk signals, tune the alert threshold, and make review decisions. All processing runs in this browser.

View stream metrics

Transactions evaluated0

Flagged for review0

Flag rate0%

Confirmed fraud (reviewed)0

Live transactions

Stream paused · synthetic data

Alert threshold: 60/100

Stream interval

View

0 transactions

Import CSV

Click any row to investigate. Import CSV columns: id, amount, type, country, accountAgeDays, hour, velocity1h, deviceChanged. Optional isFraud enables evaluation against supplied labels.

Model diagnostics

Calculated from current stream

02 / Understand the signals

Transaction investigation

Select a transaction from the stream to inspect its risk signals and review status.

01 / Experiment with risk

Test a transaction

Try a scenario, then adjust its inputs to see what changes.

Create a transaction and see how the detection engine responds.

Amount ($)

Type

Country

Account age (days)

Hour of day

Transactions in past hour

New or changed device

Recent decisions

No reviews recorded yet.

Model disclosure: This MVP uses a transparent weighted heuristic scoring engine, not a trained XGBoost classifier. Its scores are risk indices, not calibrated fraud probabilities. The stream is synthetic, and review decisions remain in the current browser session. A deployed ML inference service and persistence layer are future development stages. Related Snowflake fraud analysis ↗