Immanuel A

Versatile IT engineer with over three years of experience in infrastructure support, and automation across cloud and enterprise environments. Skilled in tools like Microsoft 365, Azure, Exchange, Intune, Freshservice, SolarWinds, and AI-based monitoring (ZIF). Experienced in implementing chatbot workflows and streamlining ITSM processes.

Now actively transitioning into the field of Artificial Intelligence, with hands-on project experience in machine learning, deep learning, Data Science and computer vision. Passionate about building intelligent systems that solve real-world problems, and continuously exploring how AI can enhance IT operations, automation, and decision-making.

Capstone Project – Digital Marketing Campaign Conversion Prediction

• Built a machine learning model to predict customer conversions from marketing data with 90:10 class imbalance using blending ensemble (Logistic Regression + XGBoost).
• Performed univariate and bivariate EDA on behavioral features (Click Through Rate, Time on Site, Email Opens/Clicks, Ads Viewed) to guide feature selection.
• Reduced input features to 12 using SelectKBest with mutual_info_classif to improve interpretability and training efficiency.
• Achieved 92% accuracy with F1-scores: 0 (No Conversion) – 0.62, 1 (Conversion) – 0.96.
• Tools and Algorithms Used:
  Python, pandas, numpy, matplotlib, seaborn, scikit-learn (Logistic Regression, SelectKBest, mutual_info_classif), XGBoost, ensemble blending, binary classification, class imbalance handling

Deep Learning Project – Safety Glass Detection Using YOLOv5

• Developed a YOLOv5-based object detection model to classify eyewear types (e.g., safety glasses, sunglasses, protected glasses, no glasses) for real-time PPE compliance monitoring in industrial settings.
• Created and annotated a custom dataset; addressed class imbalance and false positives by adding background-only images and refining annotation quality.
• Tweaked hyper parameters like IoU threshold and confidence threshold improving model accuracy.
• Deployed model for real-time inference across image, video, and webcam streams using custom OpenCV-based Python scripts.
• Achieved high performance with mAP@0.5 = 0.94 and class-wise APs: No Glass – 0.995 Glass – 0.902 Sun Glass – 0.925 Protected Glass – 0.939
• Tools and Algorithms Used: YOLOv5, PyTorch, OpenCV, LabelImg, pandas, matplotlib, Python, custom object detection pipeline, mAP/PR evaluation, real-time deployment