← thisisphume

AML Explainability Suite

A foundational client-understanding layer that captures and explains client behavior across multiple dimensions. Not a replacement for specialized models or TM rules — those plug in as add-ons. This is the base.

CWAE-MMD DynamicLRP SHAP GNN LLM-CoT FINTRAC
Interactive Mockups
Retail Client

Individual client explainability dashboard with well-studied AML typologies.

  • Contextual anomaly scoring (CWAE-MMD)
  • Feature attribution (DynamicLRP)
  • Transaction network patterns (GNN)
  • AML typology mapping + FINTRAC
  • Investigation narrative (LLM-CoT)
→ open mockup
Small Business

Industry-contextual analysis with expected vs. actual patterns and money flow.

  • Expected vs. actual by industry (NAICS)
  • Industry benchmark comparison
  • Business money flow + gap analysis
  • Revenue seasonality detection
  • SHAP + DynamicLRP attribution
  • Network pattern + typology mapping
→ open mockup
The Idea

Foundational vs. Add-On

Traditional AML: you build a model (IF, VAE, rules), then bolt on explainability (SHAP, percentiles) as a post-hoc step. Each project has different features, different approaches, different explanations. Nothing connects.

This suite flips it: explainability is the foundation. Multiple techniques (contextual scoring, feature attribution, graph reasoning, LLM narratives) work together to understand the nature of a client. Specialized models and TM rules plug in on top.

Traditional: Model → Alerts → Explainability (per project, disconnected) This Suite: Foundational Explainability Layer (multi-lens) | Understands client nature | Specialized models / TM rules plug in as add-ons
Foundational Modules
CWAE-MMD
Contextual anomaly scoring
DynamicLRP
Signed feature attribution
GNN
Network pattern detection
LLM-CoT
Investigation narrative