Sovereign AML and transaction-monitoring copilot
Ledger is a sovereign copilot for financial-crime teams. It reads flagged transactions and customer history, drafts the suspicious-transaction narrative an analyst would otherwise write by hand, and shows its reasoning, all on infrastructure the bank controls.
A flagged transaction, worked the way Ledger would: signals weighed, customer risk assessed, and a suspicious-transaction narrative drafted for an analyst to file. Runs on the bank's own infrastructure. An illustration of the workflow, not live customer data.
Drafted STR narrative
Account ****4471, opened 11 days prior, received three cash deposits of R49,500 at one branch over four days, each below the R50,000 reporting threshold, followed by a full EFT to a new beneficiary within 24 hours. Volume is inconsistent with stated monthly income of R18,000. Pattern is consistent with structuring and layering. Recommend submission to the FIC.
Reasoning trail retained for FIC and SARB
Transaction and KYC data is triple-bound: POPIA, the FIC Act, and SARB Directive 3 on cloud computing and offshoring. An offshore LLM endpoint touches all three. Ledger is designed so the model runs inside the bank or on sovereign cloud, with an explainability layer that lets a SARB or FIC reviewer see exactly why a narrative was drafted the way it was.
Flagged transactions and linked KYC records are pulled into a sovereign workspace. No data egress.
The model assembles the risk picture and proposes a narrative, citing the specific signals it used.
An analyst edits and approves. The copilot never files anything autonomously.
The reasoning trail is preserved so the decision is defensible to FIC and SARB.
Regulatory frame
Ledger is being built with design partners. If this is your problem, help shape it.
