MuleMind

Inspiration

Financial fraud and money mule networks cost the global economy hundreds of billions annually, yet most banks still rely on slow, rule-based systems that flag transactions days after the damage is done. We were inspired by the gap between the sophistication of modern financial crime and the outdated tools used to fight it. We wanted to build something that gives AML analysts a real-time, visual edge — not just alerts, but understanding.

What it does

MuleMind is a real-time Anti-Money Laundering (AML) intelligence dashboard that:

Detects mule accounts using behavioral scoring across transaction velocity, night activity, crypto exits, and fan-out patterns Visualizes money flow networks so analysts can trace how funds move through relay accounts Maps geographic risk by plotting flagged accounts on a live interactive global map Scores every account on a 0–100 risk scale across five levels: Safe, Low, Medium, High, and Critical Imports raw transaction data via CSV/Excel and instantly generates a full risk report Provides an AI assistant for querying patterns and getting natural language explanations of suspicious behavior

How we built it

Frontend: React 18 with hooks for all state management and real-time UI updates Charts & Visualization: Recharts for area, bar, and radar charts; Leaflet.js for the geographic risk map Analysis Engine: A custom-built, pure JavaScript transaction scoring engine — no external ML libraries — using weighted behavioral signals like night ratio, round amounts, crypto exits, and graph fan-out Styling: Fully custom CSS-in-JS with a dual dark/light theme system using CSS variables Data ingestion: Client-side CSV/Excel parsing so no data ever leaves the browser AI Tab: Powered by the Anthropic Claude API for natural language AML analysis

Challenges we ran into

Deterministic scoring without ML: Building a risk engine that produces meaningful, differentiated scores across five risk levels — without any trained model — required careful weight calibration and signal separation Graph rendering at scale: Drawing money flow networks with hundreds of nodes and edges without a dedicated graph library, using only Canvas and React, pushed performance limits Leaflet + React integration: Managing Leaflet's imperative DOM lifecycle inside React's declarative model required careful ref handling and cleanup to avoid memory leaks Theming consistency: Making every component — including third-party charts and the map — respect the dark/light theme system required overriding deeply nested vendor styles

Accomplishments that we're proud of

A zero-dependency analysis engine that accurately identifies mule relay patterns, structuring behavior, and crypto exits from raw CSV data in milliseconds A fully self-contained single-file React component — nearly 2,900 lines — that delivers a production-grade dashboard experience Real geographic intelligence that maps risk hotspots, offshore nodes, and cross-border money corridors on a live map A dual theme system that maintains visual consistency across every component, chart, and map layer Building something that genuinely looks and feels like a tool a real AML compliance team could use on day one

What we learned

Behavioral heuristics, when carefully combined, can rival ML models for financial anomaly detection — especially when data is limited Visualizing graph relationships is often more valuable to an analyst than raw numbers alone Performance in browser-based data analysis requires thinking carefully about when to compute, cache, and defer Designing for analysts means prioritizing information density and speed over simplicity — the opposite of most consumer UX

What's next for MuleMind

Live bank API integration — connect directly to transaction streams for true real-time monitoring Graph ML models — replace heuristic scoring with GNN-based (Graph Neural Network) mule detection trained on labeled fraud data Case management — let analysts open, annotate, escalate, and close investigations from within the dashboard SAR auto-generation — automatically draft Suspicious Activity Reports from flagged accounts using the AI layer Multi-institution collaboration — federated risk sharing across banks without exposing raw customer data Mobile alert app — push critical-risk notifications to compliance officers on the go

Built With

• css
• html
• javascript
• json
• react18