Portfolio · Compliance and Risk
Sentinel
A prototype transaction-monitoring tool. It scores customers and transactions for money-laundering risk and explains each score in plain English, so an analyst can see why something was flagged.
- Discipline
- Anti-money-laundering · Transaction monitoring
- Status
- Prototype. Runs on synthetic data only.
Request access to open it. The demo has its own sign-in and runs on synthetic data only.
The problem it solves
Compliance teams have to watch large volumes of transactions for patterns such as structuring, funnel accounts and rapid pass-through. Simple single-limit checks raise many alerts that are hard to explain. Sentinel combines several independent signals into one score and shows the reasons behind it.
How it works
01
Transactions and customer records are loaded into the engine.
02
Rules and a combined 0-100 customer risk score run over them. A separate anomaly model gives a second, independent view.
03
Customers above the alert threshold go to the alert queue with a plain-English "why flagged" and a score breakdown.
04
An analyst opens the case, reviews the timeline and contributing transactions, adds notes and escalates or closes it.
05
A draft suspicious activity report can be exported as a PDF watermarked DRAFT for a person to complete.
Key features
Risk scores with reasons
Each alert shows a plain-English reason, a score breakdown and the transactions that contributed to it.
Alert queue and case view
Alerts are queued for review. A case view shows the timeline, notes, status and an audit history.
Network view
A network chart shows how accounts connect, which helps spot rings of circular transfers and funnel accounts.
Role-based access and audit log
Analysts review, note and escalate. A compliance head can also close escalated cases and see the audit log.
Who it is for
Built as a prototype for Ugandan banks, SACCOs and forex bureaus.
What the demo shows
- A compliance overview, the alert queue, a case view, network charts and the audit log.
- A draft SAR PDF, watermarked DRAFT, with illustrative fields. It is not an official FIA format.
What is real and what is example
Real
- The scoring engine, dashboard, case workflow, audit log and draft PDF export all run in the demo.
- It was tested against a planted answer key that the engine never reads.
Example
- Every customer, account and transaction is fake: about 600 synthetic customers and 128,000 transactions.
- Results on planted, simplified data are an upper bound. Real bank data is noisier and would score lower.
- The thresholds and high-risk country list are illustrative and would need confirming with a compliance lawyer, the FIA and current FATF lists.
- Not for use on real data without a security review, calibration and data protection registration.
Screens from the demo
All data in these screens is synthetic. Detection rules and thresholds are kept confidential and are not shown.
