Deployed experiment / model serving

Live ML Demo

Manufacturing Defect Prediction — an inspectable Random Forest classifier served as a FastAPI microservice.

Evaluation snapshot

Held-out test-set metrics from this synthetic-data experiment.

Test accuracy

0.881

Train Samples

4000

Test Samples

1000

Algorithm

Random Forest

Feature Importances

temperature
44.7%
humidity
13.2%
pressure
11.1%
vibration
23.2%
rpm
7.9%

Make a Prediction

Enter manufacturing sensor values to get a real-time defect prediction.

Interpretation

What this result can—and cannot—tell us

Useful here

The interface exposes model inputs, response probabilities, and feature importances so the prediction path is visible rather than treated as a black box.

Not a deployment claim

Synthetic data cannot establish field performance. A real deployment would require representative labels, calibration, threshold selection, drift monitoring, and human review.

Technical Details

Model

Random Forest Classifier, scikit-learn, trained on 5,000 synthetic manufacturing sensor samples.

Serving

FastAPI + Uvicorn, Docker container, GitHub Actions CI/CD, deployed on Azure App Service.

Endpoint

POST /predict
Content-Type: multipart/form-data