Streaming + MLOps

Streaming Fraud Detection

A streaming analytics system that processes transaction events, engineers rolling features, scores suspicious activity, and publishes alerts for analyst review.

#Kafka#Spark#Python#MLflow#FastAPI

GitHub Live Demo

Overview

A streaming analytics system that processes transaction events, engineers rolling features, scores suspicious activity, and publishes alerts for analyst review.

Problem

Fraud operations needed lower-latency signals than daily batch reports could provide.

Solution

Combined stream processing, model scoring, and operational dashboards to surface suspicious behavior within seconds.

Architecture

How the system is structured

Kafka topics for card transaction events

Spark Structured Streaming for enrichment and rolling aggregates

Feature store tables for model-ready behavioral signals

API and dashboard layer for alert triage

Tech Stack

KafkaSparkPythonMLflowFastAPIDocker

Key Features

Sliding-window risk featuresModel registry-ready scoringAlert severity routingSynthetic event generator for demos

Screenshots

Visual walkthrough

Risk heatmap

Streaming throughput panel

Alert queue