PRODUCTION SYSTEM
Wattzo Energy Forecasting
Solar and grid demand prediction · Deployed across 12 sites
Dual-regime LSTM ensemble with conformal prediction for 12-48 hour forecasts. Reduces operator decision uncertainty for safe grid balancing.
Welcome
AI systems you can trust.
I'm a data scientist and statistician-economist building machine learning systems that combine rigorous uncertainty quantification with explainability. My research spans energy forecasting, credit risk modeling, and human-AI collaboration in high-stakes decision-making.
I work on trustworthy ML: probabilistic forecasting, conformal prediction, interpretable feature attribution, and cost-aware deferral policies that route uncertain predictions to humans. Recently completed my M.Sc. in Artificial Intelligence for Science at AIMS South Africa (Google DeepMind Scholar), with a thesis on explainable probabilistic forecasting for energy systems.
I'm open to roles in: trustworthy AI, uncertainty quantification, explainable ML, energy systems, and ML engineering. Currently considering opportunities in AI research labs and industry ML teams, with flexibility for remote or relocation after August 2026.
Recent work
POLICY DASHBOARD
SME Survival Early-Warning
Econometric + ML pipeline · Cameroon Ministry deployment
Hybrid CatBoost and survival model identifying high-failure-risk enterprises 12–24 months ahead. Targets government support to protect jobs.
PEER-REVIEWED
SemEval-2026 NLP Competition
5th place (Hausa) · ACL proceedings
Multilingual transformer with label-wise optimization for polarization detection in low-resource languages. Published in top-tier venue.
What I've learned
On research
Uncertainty quantification isn't an afterthought—it's foundational. Conformal prediction and Bayesian approaches scale better than post-hoc calibration. Explainability through feature attribution (SHAP) and uncertainty decomposition earns trust from operators and policymakers.
On deployment
Real systems need deferral strategies. Not every prediction should be automated—humans should route uncertain cases. Cost-aware thresholds and fairness constraints beat simple probability cutoffs.