End-to-End MLOps Pipeline for Credit Risk Monitoring

Production-style MLOps pipeline for credit risk prediction, model serving, drift detection and observability.

Overview

This project simulates an end-to-end machine learning system in a production environment, covering the full lifecycle from model training to deployment and monitoring.

It includes:

• Predictive model for credit behavior
• FastAPI model serving API
• Dockerized deployment
• Data drift monitoring framework
• Interactive Streamlit observability dashboard
• Deployment in Render (API + Monitoring Console)

The project was designed following MLOps principles: reproducibility, monitoring, model governance and operational visibility.

---

Architecture

Data → Feature Engineering → Model Training → API Inference
                                     ↓
                           Drift Monitoring Layer
                                     ↓
                           Monitoring Dashboard

Components

1. Model Training

• Feature engineering pipeline
• Classification model for Pago_atiempo
• Serialized production pipeline with Joblib

Algorithms explored:

• Logistic Regression
• Gradient Boosting approaches
• Risk-oriented evaluation metrics

---

2. Prediction API (FastAPI)

Deployed API:

https://mlops-pipeline-api.onrender.com

Endpoints:

Health Check

GET /

Batch Predictions

POST /predict

Returns:

• predicted class
• probability score
• model version
• threshold metadata

Example response:

{
 "predictions":[
   {
    "proba_pago_atiempo":0.983,
    "pred_pago_atiempo":1
   }
 ]
}

Swagger Docs:

https://mlops-pipeline-api.onrender.com/docs

---

3. Drift Monitoring Console

Live dashboard:

https://mlops-pipeline-dashboard.onrender.com

Monitoring includes:

Numerical drift:

• PSI
• Kolmogorov-Smirnov
• Jensen-Shannon Distance

Categorical drift:

• Chi-Square Drift Detection

Dashboard modules:

• Executive Risk Overview
• Drift Incident Panel
• Feature Diagnostics
• Historical Drift Monitoring
• Risk Scoring Matrix
• Downloadable Monitoring Reports

---

Monitoring Example

Detected incident:

• Feature: tendencia_ingresos
• Status: DRIFT
• Automatic recommendation: Review upstream category mapping

Simulates production monitoring workflows.

---

Tech Stack

Python
Pandas
Scikit-learn
FastAPI
Streamlit
Docker
Render
Joblib
Plotly

---

Project Structure

mlops_pipeline/
│
├── data/
├── models/
├── notebooks/
├── reports/
├── src/
│   ├── app.py
│   ├── model_deploy.py
│   ├── model_monitoring.py
│   └── ft_engineering.py
│
├── Dockerfile
├── docker-compose.yml
└── requirements.txt

---

Local Run

Clone repository:

git clone https://github.com/Florenciasc/mlops_pipeline.git
cd mlops_pipeline

Install:

pip install -r requirements.txt

Run API:

uvicorn src.model_deploy:app --reload

Run monitoring dashboard:

streamlit run src/app.py

---

Key MLOps Capabilities Demonstrated

✔ Model deployment
✔ Batch inference API
✔ Data drift detection
✔ Model observability
✔ Incident-style monitoring
✔ Dockerized deployment
✔ Cloud deployment
✔ Monitoring dashboard

---

Future Improvements

• Model performance monitoring (prediction drift)
• Automated alerting
• CI/CD pipeline integration
• Retraining triggers
• Prometheus/Grafana integration

---

Author

Florencia Sosa Comisso
Business & Data Analyst | MLOps | BI

LinkedIn: https://linkedin.com/in/florencia-sosa-comisso

Portfolio: https://florenciasc.github.io/