Real-time
Feature Engineering
Streamlit
Monitoring UI
Anomaly
Detection
0
Data Leakage
Real-time feature engineering pipeline for fraud detection, with a Streamlit dashboard for monitoring fraud scores and anomaly patterns. Demonstrates streaming feature computation, point-in-time correctness, and production fraud scoring patterns.
Real-time feature engineering done right: point-in-time correctness, no data leakage, streaming patterns.
Tech Stack
PythonStreamlitscikit-learnPandasFastAPIDuckDBPlotly
Features
Streaming Features
Real-time feature computation with point-in-time correctness. No future data leakage into training.
Fraud Scoring
ML model scores each transaction in real-time with interpretable feature contributions.
Anomaly Patterns
Streamlit dashboard visualizes fraud pattern clusters, score distributions, and detection rates.
FastAPI Serving
REST endpoint for real-time scoring with sub-100ms latency on standard hardware.
How It Works
01
Stream
Transaction events → real-time feature computation engine
→
02
Features
Point-in-time correct features: velocity, pattern, behavioral signals
→
03
Score
FastAPI → ML model → fraud score + feature importance
→
04
Monitor
Streamlit dashboard → score drift, anomaly clusters, detection rate