Birrdash

AI-powered fraud detection for global remittances-stopping $40B in fraud with real-time risk scoring,KYC verification,and sanctions screening. Built on LiquidMetal's Raindrop Platform, Claude & Vultr.

Comment

Inspiration

The global remittance market moves $800 billion annually, yet operators lose more than $40 billion to fraud, leakage, sanctions violations, and compliance failures every year. Traditional fraud detection systems are slow (3–5 minutes), reactive, and unable to detect modern threats like synthetic identities, mule networks, or corridor-level fraud.

We saw an opportunity to build something fundamentally different:

Real-time, AI-powered fraud detection that evaluates risk before transfers are executed. By combining advanced analytics with LiquidMetal’s Raindrop Platform, Claude’s code generation, and Vultr’s cloud infrastructure, we built BirrDash — a fast, accurate, and highly accessible governance layer designed specifically for licensed remittance operators.

BirrDash (https://www.birrdash.com) is a real-time fraud detection and sanctions screening platform for the global remittance industry. It helps operators make safe payout decisions by delivering instant risk insights and actionable recommendations.

Core Features Multi-Factor Fraud Detection BirrDash combines multiple signals into a unified risk score:

Velocity checks Transaction amount anomaly detection Corridor & geolocation patterns Device fingerprinting IP reputation Sanctions screening

Behavioral indicators KYC Verification Automated identity checks Politically Exposed Person (PEP) flagging Identity consistency scoring

Sanctions & Watchlist Screening Real-time checks against: OFAC UN sanctions EU + global watchlists

Interactive Fraud Dashboard (6 Tabs)

Operators can: Search 1,003 customers Analyze 10,000 transactions Run fraud simulations Inspect behavioral and risk indicators Conduct sanctions reviews View instant risk scoring

Lightning-Fast Detection (<100ms) BirrDash processes and returns a risk verdict in under 100 milliseconds — compared to 3–5 minutes for traditional batch-based fraud engines.

Users receive: A complete risk breakdown Recommendations (Approve / Escalate / Block) Explanations of risk drivers

How we built it

Frontend Pure Vanilla JavaScript No framework bloat Realtime charts, search, and dynamic risk UI Lightweight, operator-friendly design

Backend Node.js + Express API JWT authentication Highly optimized risk-score endpoint Modular rule-based and ML-ready fraud logic

Database Vultr Managed PostgreSQL Seeded with 1,003 customers + 10,000 transactions Indexed for fast lookups Structured for analytics expansion

AI Integration Claude Code for rapid prototyping and API construction Raindrop SmartSQL for natural-language-driven database queries AI-assisted rule building and fraud scenario testing

Deployment Vultr Cloud Compute (Ubuntu) PM2 process management Secure environment variables Configured for scalable read/write concurrency

Challenges we ran into

  1. MCP SDK ES Module Compatibility: Hit require() of ES Module errors with @modelcontextprotocol/sdk. Solution: Created simplified API for demo while maintaining MCP in templates.
  2. Production Deployment: Server confusion between local dev and Vultr production. Learned to verify working directory and PM2 process management.
  3. Browser Caching: Dashboard showed old UI despite correct deployment. Fixed with hard refresh education.
  4. Credential Security: Discovered exposed database passwords in git history. Pivoted to ZIP submission and planned password rotation.

Accomplishments that we're proud of

Speed of Development: Built a production-ready fraud detection platform using Claude Code - from zero to 10,000 seeded transactions with full dashboard and API Performance Breakthrough: Achieved 2,000x faster fraud detection (<100ms) compared to traditional batch processing systems (3-5 minutes) Real Impact: Successfully flagged 2,890 high-risk transactions from our test dataset, demonstrating effective fraud pattern recognition Production-Grade Architecture: Deployed on live Vultr infrastructure with managed PostgreSQL Scalable API handling concurrent requests Professional dashboard with JWT authentication 973 lines of vanilla JavaScript - no framework bloat AI-Powered Development: Leveraged Raindrop Platform's SmartSQL MCP for natural language database queries and Claude Code for rapid prototyping - proving AI-assisted development works at scale

What we learned

Claude Code acceleration: AI-assisted development cut coding time by 70% - from database design to API implementation in hours, not days Raindrop Platform power: SmartSQL MCP server enabled direct PostgreSQL queries with natural language, eliminating boilerplate Vultr infrastructure: Managed PostgreSQL and Object Storage provided production-grade reliability without DevOps overhead Real-world fraud patterns: Velocity checks, geolocation analysis, and sanctions screening must work together for effective detection

What's next for Birrdash

Immediate Roadmap (Q1 2026) Advanced ML Models: Train custom fraud detection models on real-world remittance data to improve accuracy beyond rule-based scoring Mobile SDK: iOS and Android libraries for embedded fraud detection in remittance apps Real-Time Alerts: WebSocket-based notifications for high-risk transactions with configurable thresholds

Medium-Term Goals (Q2-Q4 2026) Multi-Currency Support: Expand beyond USD to handle EUR, GBP, ETB, and 50+ currencies with region-specific fraud patterns Partner Integrations: Direct API connections with major remittance providers (Western Union, MoneyGram, Wise) Behavioral Analytics: Track user patterns over time to detect account takeovers and identity theft Banking-as-a-Service: White-label fraud detection platform for banks and fintech startups Blockchain Integration: Verify cross-border transactions on distributed ledgers for immutable audit trails Global Expansion: Support 25+ countries with localized compliance rules and regulations

Built With

Share this project:

Updates