● Production ML· Finance·2025Shipped
FraudGuard-AI
ML fraud detection dashboard — XGBoost classifier behind FastAPI, real-time transaction scoring, Chart.js visualizations.
Overview
A production-style fraud detection dashboard. Transactions stream into a FastAPI endpoint, an XGBoost classifier scores each one for fraud likelihood, and a live dashboard surfaces flagged transactions with the model's confidence and feature attributions.
Architecture
- FastAPI serves the inference endpoint with sub-100ms per-request latency
- XGBoost classifier trained on a labeled transaction dataset with class imbalance handled via SMOTE
- Chart.js dashboard renders a live risk-score histogram and a transaction stream with flag markers
- Feature attribution per flagged transaction so analysts can see why the model flagged it
What it shows
This project shifts from notebook-style analysis to a deployable system — the parts that matter when ML actually goes into production: API design, latency budgets, threshold tuning for the imbalanced positive class, and a UI a non-ML user can actually act on.
Stack
- Python · XGBoost · FastAPI for the backend
- CSS · Chart.js for the dashboard
- Docker for deployment