I'm Nuru Nabuuso. I build AI systems whose reasoning can be seen, audited, and trusted — and I apply them to high-stakes health problems, starting with childhood malnutrition in Uganda.
Powerful AI is spreading into clinics, banks, and ministries. Most of it cannot explain itself — and a decision no one can question is a decision no one can trust.
I'm a Ugandan PhD researcher in Explainable Artificial Intelligence at Universitat Pompeu Fabra in Barcelona, supervised by Prof. Vladimir Estivill-Castro.
My work designs AI systems whose decision-making is understandable by construction — not explained after the fact by external tools, but transparent from the inside out. The goal is that clinicians, policymakers, and the people affected by a model can interpret, validate, and act on its predictions with confidence.
I came to this through health informatics. Before my PhD I earned a Master's at Makerere University and a computing degree at Kyambogo University. That background keeps my research anchored to a real question: would a health worker actually trust and use this?
Sparse oblique decision trees and forests that stay auditable as they scale — interpretability that originates in the model, not a wrapper around it.
Applying transparent ML to childhood malnutrition and community health screening, validated against national survey data.
Studying what makes an explanation usable by a real clinician in a low-resource setting — where compute, data, and trust are all scarce.
A human-centred framework for building oblique decision trees that remain interpretable — using parallel-coordinate visualisation and sparsity to keep expressive splits auditable, where traditional oblique trees become opaque.
Addresses the interpretability of ML in high-stakes domains: a greedy approach to sparse linear models that preserves accountability and trust where black-box accuracy alone undermines ethical and legal compliance.
Compares ML models — SVM, XGBoost, and neural networks — for predicting child malnutrition in Uganda, and evaluates how well LIME and SHAP explain their decisions. XGBoost gave the strongest predictive performance.
A six-input mobile app for community health workers that flags childhood stunting risk. Runs fully offline in the browser after first load — the model executes on-device, no internet needed for a prediction.
A full stunting-risk predictor with SHAP explanations in plain language — showing exactly which factors drive each prediction. Built for researchers, clinicians, and District Health Officers, with a live population view.
An explainable AI lab in Uganda — building trustworthy AI for health and other high-stakes domains, and training the researchers who will carry it forward.
Most health AI deployed across Africa is trained on data that doesn't represent the people it serves, and offers no way to question its outputs. I want to change where that work happens — and who does it.
The long-term ambition is a research lab rooted at home: studying interpretable AI, working directly with Makerere University and Uganda's Ministry of Health, and proving that transparency, local relevance, and strong performance can hold together at once.
Universitat Pompeu Fabra, Barcelona · Department of Engineering. Interpretable models for high-stakes decision-making.
Makerere University, Kampala. Where the health focus and the malnutrition research began.
Kyambogo University, Kampala.
I'm open to research collaborations, institutional partnerships in Uganda, funding conversations, and speaking. If you work in explainable AI, global health, or AI for social good — get in touch.