Fraudies

Overview

A modular, full-stack fraud detection system combining deterministic velocity rules with a machine learning scoring engine to assess transaction risk in real time. Built as a portfolio demonstration of enterprise fraud-monitoring architecture patterns.

Fraudies is a three-service fraud detection system: a NestJS REST API that handles ingestion and orchestration, a FastAPI ML inference engine that scores transaction risk, and a Next.js analyst dashboard for real-time monitoring. PostgreSQL stores all transaction records and audit history. Redis handles velocity-based rule checks.

Key Features

Technology Stack

Challenges

• Achieving sub-50ms end-to-end latency for real-time fraud scoring

• Balancing false positives vs. false negatives in ML model tuning

• Building immutable audit trail with database-level enforcement

• Isolating ML service from direct database access for security

• Creating intuitive risk factor explanations from ML predictions

Solutions

• Implemented three-layer scoring: Redis velocity checks, ML inference, threshold classification

• Used feature importance extraction to generate human-readable risk factors

• Enforced audit immutability with PostgreSQL triggers + service-layer constraints

• Designed ML engine as stateless service with no DB credentials

• Built WebSocket real-time dashboard updates with subscription patterns

Project Architecture

fintech-fraudies/
├── frontend/             # Next.js analyst dashboard
│   ├── app/
│   │   ├── (auth)/      # Login and registration
│   │   ├── dashboard/   # Main analyst view
│   │   ├── transactions/ # Transaction list/detail
│   │   └── alerts/      # Flagged transaction queue
│   └── components/
│       ├── ui/          # Shared primitives
│       ├── TransactionCard/
│       ├── RiskScoreMeter/
│       └── MetricsPanel/
├── backend/
│   ├── api/             # NestJS REST API
│   │   └── src/
│   │       ├── modules/
│   │       │   ├── auth/        # JWT + bcrypt
│   │       │   ├── transactions/ # Ingestion + ML orchestration
│   │       │   ├── webhook/     # Webhook ingestion
│   │       │   └── audit/       # Append-only audit log
│   │       └── common/
│   └── ml-engine/       # FastAPI inference service
│       └── app/
│           ├── model/   # Random Forest classifier
│           └── schemas.py
└── database/
    ├── schema.prisma
    └── migrations/

Key Learnings

• Production ML deployment patterns with FastAPI and Scikit-Learn

• Microservices security architecture and service isolation

• Database-level audit enforcement with PostgreSQL triggers

• Real-time data streaming with WebSocket subscriptions

• Feature engineering for fraud detection use cases

• JWT authentication flows in NestJS with Passport.js

Future Enhancements

Async job queue with BullMQ for decoupled scoring, Prometheus metrics export, OpenTelemetry tracing, PagerDuty alert integration, and read replicas for analytics queries.