Skip to content
Shonin
Beta

AML Alert Triage

AML alerts triaged and suspicious-transaction reports drafted, with an analyst deciding every filing.

The split

LLM writes
Drafts STR narratives
System One decides
Risk score; close or escalate
Code executes
Assembles transactions, computes amounts and velocities, logs

Engine

Code assembles each alert's transactions and customer profile; the brain scores risk and decides close or escalate; an LLM drafts the STR narrative for an analyst to approve.

Rulebook

What correct means here, and the mistakes we've already caught. It grows one job at a time.

  1. R01The model never files a report; an analyst approves every STR.
  2. R02Every closure cites the transactions and the rule that clears it.
  3. R03Amounts, dates and velocities are computed in code, never by a model.
  4. R04Model version and confidence are logged per alert for the examiner.

Intake

  • Alert export and rule definitions
  • Customer risk tiers
  • 500+ past dispositions
  • Your STR template
UNIT
One alert triaged
REVIEW LAYER
Analysts review every escalation and a random 10% of closures.
DELIVERY
Dispositions and STR drafts in your case manager, with an audit log.
TURNAROUND
Same day

How we find buyers

  • Compliance leads at fintechs facing the CBN's automated AML requirement
  • Case-management vendors as partners
  • Agent Reliability Audit clients

More like this