AI JOB SEARCH AUTOMATION

"One project I'm particularly proud of is my AI Industrial Inspection project, which is a computer vision-based deep learning application for detecting defects on steel surfaces.

The main objective was to classify steel surface images into 10 different types of industrial defects. For this, I used transfer learning with EfficientNet-B0 using PyTorch.

I worked on the complete pipeline, starting from image preprocessing and model development to evaluation and deployment. The model achieved around 87.86% test accuracy and 79.61% macro F1 score on the held-out test data.

One part I found particularly interesting was explainability. I integrated Grad-CAM so that, along with the predicted defect and confidence score, the application could also show which regions of the image were influencing the model's prediction.

Finally, I deployed the application using Streamlit, where a user can upload a steel surface image and get the predicted defect, confidence score, Top-3 predictions and Grad-CAM visualization. So overall, it gave me experience in taking a deep learning model from development all the way to a usable application."