metadata
title: Fraud Guard Intelligence
emoji: 🛡️
colorFrom: blue
colorTo: indigo
sdk: streamlit
sdk_version: 1.57.0
app_file: app/streamlit_app.py
pinned: true
license: mit
💳 Fraud Guard Intelligence: Adaptive MLOps System
This repository hosts the interactive frontend for a production-ready Fraud Detection system. The architecture is designed to handle non-static fraud patterns through real-time inference, explainable AI, and a human-in-the-loop retraining pipeline.
🚀 System Architecture
This is a Microservices-based Deployment:
- Frontend (This Space): Streamlit UI for transaction input and SHAP-based risk visualization.
- Backend (Render/Railway): FastAPI server handling high-concurrency inference requests.
- Data Layer (Neon DB): PostgreSQL cloud database logging real-time telemetry for drift analysis.
- Experiment Tracking (DagsHub/MLflow): Versioned model registry and performance tracking.
🧠 Key Features
- Explainable AI: Uses SHAP Waterfall plots to visualize why a specific transaction was flagged.
- Dynamic Sensitivity: Stakeholders can adjust the "AI Sensitivity" slider to balance False Positives vs. False Negatives.
- Drift Monitoring: Integrated with Evidently AI to detect statistical shifts in incoming data distribution.
- Auto-Retraining: Trigger a model refresh directly from the UI when enough human-verified data is collected.
🛠️ Tech Stack
- Languages: Python
- ML Frameworks: Scikit-Learn, XGBoost (v2.0.3)
- APIs: FastAPI, Uvicorn
- Tracking: MLflow, DagsHub
- Database: SQLAlchemy, PostgreSQL (Neon)
- Monitoring: Evidently AI, SHAP
🔐 Environment Setup
To run this project locally or on your own Space, ensure the following Secrets are configured:
DATABASE_URL: Cloud PostgreSQL connection string.API_URL: The URL of your deployed FastAPI backend.DAGSHUB_USER_TOKEN: For MLflow logging.MLFLOW_TRACKING_URI: DagsHub MLflow remote URI.
📂 Project Structure
app/: Contains the Streamlit dashboard logic.src/: Modular code for pipelines (Predict, Train, Drift).artifacts/: Serialized model files (.pkl) and scalers.app.py: The entry point for Hugging Face deployment.
Author: Mohit Parmar
GitHub: MohitParmar78