FraudNet.AI

Real-time financial fraud detection system with sub-200ms inference, 94.2% accuracy, and production-grade security.

FraudNet.AI is a production-grade financial fraud detection platform that combines machine learning with enterprise security. Built with Flask and XGBoost, it processes transactions in real-time with sub-200ms inference latency while maintaining 94.2% accuracy and ~90% F1 score. The system features SMOTE and class-weight tuning for imbalanced datasets, JWT + RBAC authorization (Admin / Analyst / Viewer), Redis-backed sessions with rate limiting, and append-only audit logs for compliance. Deployed via Docker + Nginx with Prometheus + Grafana monitoring, CI/CD via GitHub Actions, and 95%+ test coverage across 450+ tests.

Tech Stack

PythonFlaskXGBoostNext.jsMySQLRedisKafkaDocker

Features

Sub-200ms real-time fraud inference on financial transactions
94.2% accuracy with ~90% F1 score using XGBoost
SMOTE + class-weight tuning for imbalanced fraud datasets
JWT authentication with RBAC (Admin / Analyst / Viewer roles)
Redis-backed session management with rate limiting
Append-only audit logs for regulatory compliance
Docker + Nginx production deployment
Prometheus + Grafana observability and monitoring
CI/CD pipeline via GitHub Actions
95%+ test coverage with 450+ automated tests

Challenges

Achieving sub-200ms inference latency under production load.
Handling severely imbalanced fraud datasets with SMOTE and class-weight tuning.
Implementing multi-role RBAC authorization with secure JWT sessions.
Building append-only audit logging for compliance requirements.
Maintaining 95%+ test coverage across 450+ tests with CI/CD enforcement.

Learnings

Designed low-latency ML inference pipelines for real-time decision systems.
Implemented production-grade security patterns (JWT, RBAC, rate limiting).
Built comprehensive observability with Prometheus and Grafana.
Developed robust CI/CD workflows with GitHub Actions for automated testing and deployment.

Feedback

For feedback or suggestions, contact me at: shouriyatayal1234@gmail.com

Link

Github

FraudNet.AI

Real-time financial fraud detection system with sub-200ms inference, 94.2% accuracy, and production-grade security.

Features

Sub-200ms real-time fraud inference on financial transactions

94.2% accuracy with ~90% F1 score using XGBoost

SMOTE + class-weight tuning for imbalanced fraud datasets

JWT authentication with RBAC (Admin / Analyst / Viewer roles)

Redis-backed session management with rate limiting

Append-only audit logs for regulatory compliance

Docker + Nginx production deployment

Prometheus + Grafana observability and monitoring

CI/CD pipeline via GitHub Actions

95%+ test coverage with 450+ automated tests

Challenges

Achieving sub-200ms inference latency under production load.

Handling severely imbalanced fraud datasets with SMOTE and class-weight tuning.

Implementing multi-role RBAC authorization with secure JWT sessions.

Building append-only audit logging for compliance requirements.

Maintaining 95%+ test coverage across 450+ tests with CI/CD enforcement.

Learnings

Designed low-latency ML inference pipelines for real-time decision systems.

Implemented production-grade security patterns (JWT, RBAC, rate limiting).

Built comprehensive observability with Prometheus and Grafana.

Developed robust CI/CD workflows with GitHub Actions for automated testing and deployment.