Visualisation: a suspended lattice of steel nodes and rods in a dark concrete shaft, one path through it traced in cold cyan light.

Promeing ConceptScenario set across global markets

Risk Intelligence

IndustryFinancial services

Visualisation

Status
Promeing Concept
Scenario
Global
Industry
Financial services
Capabilities
AI AgentsAI InfrastructureAI Products
Sheet
06 / 06

Agents that link filings, transactions, sanctions lists and news into one entity graph, where every flag carries an evidence trail an analyst can audit.

A Promeing Concept: a studio study of a system we would build, set in a realistic scenario. It describes no client, no deployment and no measured result.

Problem

A risk team monitors counterparties across filings, transactions, sanctions lists and news, in several languages and jurisdictions. The signals exist, but they sit in different systems, and connecting them is manual work that is hard to repeat and harder to explain to an auditor.

Context

The scenario: the counterparty risk function of a financial services firm working across jurisdictions. Ownership structures are layered, names are transliterated differently from one source to the next, and every decision to flag or clear a counterparty has to be defensible months later, to people who were not in the room.

Visualisation: close detail of a steel node where rods meet, one of the rods traced in a cold cyan line.
VisualisationFig. 01

System

Agents extract entities and relationships from filings, registries, transaction data, sanctions lists and news, resolve them to one entity even when names are spelled differently, and link them into a graph. When a pattern meets a written rule, the system raises a flag with its evidence trail: each hop, each source document and when it was retrieved. An analyst accepts, rejects or asks for more evidence, and every decision is written to an audit log. Governance and model monitoring are built in from the first release, and a held-out set of historical cases gates every model change.

AI architecture

Sources are extracted and resolved into one entity graph. Rules and agents raise flags with evidence; the analyst decides, and the audit log keeps the decision and everything it rested on.

AI architecture of Risk Intelligence

Layers

  • Interface: Analyst workbench
  • Agents: Monitoring agents, Rules and flagging
  • Retrieval: Evidence retrieval
  • Models: Entity resolution, Extraction models
  • Data: Entity graph, Filings, transactions, lists, news
  • Infrastructure: Audit log, Evaluation gate

Connections

  • Filings, transactions, lists, news → Extraction models
  • Extraction models → Entity resolution
  • Entity resolution → Entity graph
  • Entity graph → Monitoring agents
  • Monitoring agents → Rules and flagging
  • Evidence retrieval → Rules and flagging
  • Entity graph → Evidence retrieval
  • Rules and flagging → Analyst workbench
  • Analyst workbench → Audit log
  • Evaluation gate → Entity resolution
ArchitectureWhat people use at the top, what it stands on at the bottom

Interface

The analyst workbench: the entity graph around a flagged counterparty, the evidence chain that raised the flag, and the decision that goes on record.

Interface mockSchematic, with sample data

  1. A: The entity graph: names resolved to one entity, relationships drawn from the sources.
  2. B: The evidence chain behind the flag, hop by hop, each with its source document and date.
  3. C: The analyst decides; the decision and everything it rested on go to the audit log.

Numerals appear only in dates, IDs and timestamps. Indicators read — until there is something true to show.

Technology

  • Extraction from filings, registries and news in several languages
  • Entity resolution across transliterations and spelling variants
  • Graph database for entities and relationships
  • Rules engine for flags, written and versioned with the risk team
  • Evidence retrieval with source snapshots and retrieval timestamps
  • Model monitoring and a held-out evaluation set that gates every change
  • Immutable audit log with role-based access

Expected outcome

Expected

Earlier, better-explained risk signals, analysts who spend their time on judgement rather than on joining data, and a complete audit trail for every decision.

Qualitative and unmeasured. A pilot measures it against acceptance criteria agreed at the start.

What we would measure

  1. Analyst time per case
  2. Signal precision
  3. Completeness of the audit trail
  4. Entity resolution accuracy
  5. Time from source update to flag

Metric names only. Values exist once a pilot measures them.

What a pilot would need

  1. A defined counterparty segment and the sources the team already uses
  2. Historical cases, with their outcomes, for the held-out evaluation set
  3. The current flagging rules, written down with the risk team
  4. Access controls and audit requirements from compliance
  5. Two analysts to review every flag during the pilot
  6. An environment that meets the firm’s data residency rules

Promeing ConceptScenario set across global markets

Building something like this?

Bring the problem in its messiest form. We’ll tell you honestly whether AI is the right way to solve it.

Discuss a system like this All work