📍 Kampala, Uganda  ·  🖥️ Portfolio  ·  💼 LinkedIn  ·  ✉️ Email

I build data and ML systems that hold up outside the notebook from T-SQL medallion warehouses to fairness audited clinical classifiers trained on real hospital data. My recent focus is applying rigorous MLOps (versioned data, tracked experiments, monitored deployments) to healthcare problems in resource-constrained settings, most recently a hypertension risk model built on records from Nakaseke Hospital, Uganda, evaluated for both predictive performance and demographic fairness.

• 🚀 Currently building Hypertension Risk Screener This is a fairness-audited, SHAP-interpretable classifier deployed at nakaseke-hypertension-screener.onrender.com
• 📚 Deepening my medical-imaging skills through Project MONAI's tutorials
• 👥 Open to collaborating on data engineering or clinical AI projects
• 💬 Ask me about missing-data imputation (MICE/MissForest), medallion architecture, or ML fairness audits

Hypertension Risk Screener I built an end to end MLOps pipeline for hypertension risk screening using data from Nakaseke hospital Uganda. After discovering the diabetes labels were too imbalanced for reliable learning, I reframed the problem as hypertension classification. I developed data cleaning and feature engineering pipelines, created clinically meaningful features such as BMI and Waist to hip ratio risk, benchmarked six machine learning algorithms using stratified cross validation and SMOTE, optimized th top models with Optuna, tracked experiments in MLflow, explained predictions with SHAP, and deployed th final Logistic regression model through a Flask web application containerized with Docker. scikit-learn MLflow SHAP Docker Live demo →

Kidney CT Classifier I built a complete MLOps-based computer vision system for kidney tumor detection using CT scan images. The solution uses transfer learning with VGG16 to classify scans as Normal or Tumor, achieving around 89.9% validation accuracy. Beyond model training, I implemented DVC for data versioning, MLflow and DagsHub for experiment tracking, Flask for serving predictions, Docker for containerization, GitHub Actions for CI/CD, and deployed the final application on Hugging Face Spaces. The project was designed to demonstrate a production-ready machine learning workflow rather than just a standalone deep learning model. PyTorch DVC MLflow Hugging Face Space →

Dental OPG X-Ray Analysis I built an end-to-end MLOps computer vision system for dental OPG X-ray analysis. The project uses deep learning to automatically identify dental abnormalities from panoramic radiographs and was designed as a production-grade solution rather than a standalone model. The work includes data preparation, model training, experiment tracking, reproducible pipelines, Docker-based deployment, and a web-based inference application. The objective was to demonstrate how Al can support dentists by providing rapid and scalable radiographic screening while showcasing real-world MLOps engineering practices. YOLOv8 Computer Vision MLOps

Medallion Data Warehouse I built a modern SQL Server data warehouse that consolidates data from ERP and CRM systems into a unified analytics platform. The solution uses a Medallion Architecture with Bronze, Silver, and Gold layers, implements ETL pipelines for data integration and cleansing, and models business data using star-schema fact and dimension tables. The project demonstrates end-to-end data engineering capabilities, including data warehousing, dimensional modeling, data quality management, and analytics-ready reporting. T-SQL SQL Server ETL Project site →