Oluso reads every transfer as a live graph of accounts. It spots a mule ring while the money is still moving, drafts the case with an AI investigator, and leaves the decision to a named officer.
₦850,000
gone before breakfast.
A cloned voice of her son asks for hospital money. Mrs A sends it.
It lands in an account opened six days ago and splits five ways in seconds.
Mules pass it on. One cashes out ₦170,000 at a POS agent.
She calls the bank. The money is gone, and no single transfer broke a rule.
An illustrative case. Every account and amount on this page is synthetic.
Rules that score one account at a time miss mule rings, because each account looks normal on its own. Oluso looks at the shape of the flow instead.
Every transfer updates a live graph of accounts. Each event is scored, and suspicious accounts are grouped into a ring.
An AI agent with read-only tools drafts the case in plain language, links every claim to a transaction, and proposes one action.
A named officer approves or rejects. Policy enforces hard limits, and every step goes on a tamper-evident log.
Most fraud tools add another classifier and another dashboard. Oluso adds two things banks have not had.
Temporal graph scoring and community detection find networks that per-account rules miss.
Every sentence the agent writes links to a transaction an analyst can check in seconds.
Deny-by-default tools, limits enforced in code, and a human approval for every action.
Planted rings hide among look-alike merchants, payroll runs and collections, so precision is measured, not assumed.
Nothing else touches core banking. Data stays inside the bank, accounts carry pseudonymous IDs, and no names reach the model.
Build targets, tested on labelled synthetic data. Not results yet.
Team lead: ML, agent and the pitch
Product, demo and infrastructure
The live demo plays the Mrs A case in real time on synthetic data. You make the officer's call.
Open the live demo