Marqib monitors trades, deposits and withdrawals overnight within the firm's own environment. Explainable low-risk alerts are closed with a written rationale; the remainder reach the MLRO as complete case files: evidence, reasoning and a regulator-ready draft. No report is filed without a named reviewer's approval.
Alerts are produced by rules and statistics the firm can read, tune and evidence to an auditor. Where a language model is used, it writes only from the evidence the rules have already collected, and every sentence it produces must cite a record.
End-of-day trades, orders, deposits, withdrawals and KYC changes are received from the trading platform, CRM and payment providers. File-based delivery at the outset; API connectors as required.
Typology rules run against the full book. Each hit is scored on volume, profile consistency, screening results and history. Low-risk, explainable hits are closed automatically with a recorded rationale.
For each open alert the platform assembles the linked records: beneficial ownership, funding sources, counterparties, rebate ledgers, screening results, on-chain exposure. Read-only access under the firm's existing permissions.
A hashed evidence pack and a narrative in the FIU's format, or an explicit recommendation not to report with the proposed action. Each statement in the draft references the record that supports it.
The MLRO works from a risk-ranked queue: approve, edit, request information or close as a false positive. Holds on withdrawals are placed and released from the same case file.
Every action, by the platform or by a reviewer, is written to an append-only, hash-chained log. The record of why a matter was, or was not, reported exists before the regulator asks.
Each typology is delivered with calibrated thresholds, a reasoning template and the applicable filing format. Firm-specific rules are added within the same framework.
Regulated firms are right to ask what leaves their environment and what a model is allowed to decide. The answers are designed into the platform rather than left to policy.
Whether a matter is flagged is determined by rules and statistics that the firm can inspect and tune. A language model never decides what is suspicious.
Where enabled, a model rewrites the reasoning and the report narrative from the same evidence table the rule produced. Every sentence must cite a record; sentences without a valid citation are removed before the reviewer sees them. Rule text and model text are shown side by side.
Before any model call, client identifiers, names, account numbers, card and wallet references are replaced by tokens and restored only after the response is received. The record of what was sent is auditable.
The same platform operates with a model endpoint in the firm's region (for example Azure OpenAI in the UAE or AWS Bedrock in Bahrain), with an open-weights model on the firm's own servers, or with no model at all, in which case the rule-generated draft is the narrative.
Provider, model, prompt hash and the number of pseudonymised tokens are written to the audit log for each draft.
The detection engine is common to all deployments; the filing adapter and typology thresholds follow the jurisdiction. Adapters marked as live are in use; others are scheduled with the first deployment in that market.
Marqib is installed in the firm's environment and operated alongside the existing process for one quarter. Three measures are agreed in advance and reported at the end: reviewer hours per alert, STR turnaround, and matters identified that the existing process did not.
Fixed pilot fee
Credited against the first-year licence on continuation.
Marqib is led by Anıl Abbak, who spent eighteen years in licensed brokerage: Managing Director for international markets at a leading Turkish investment house; Deputy Chief Executive of a listed investment holding; Chairman of its United Kingdom, Montenegro and Mauritius subsidiaries; and an FCA-approved Senior Manager (SMF3). He has held the review responsibility the platform is designed to support.
The platform is developed by Nordera, with delivery partners in Dubai and Istanbul.