Portfolio
Portfolio
Senior data science and AI/ML engineering work across healthcare AI, pricing, recommender systems, and experimentation.
A focused view of work that best represents my profile: production-minded machine learning, applied research, data products, and business-facing analytics with measurable impact.
Featured case studies
High-impact applied work
Causal Foundry: claims analytics and health financing models
Built healthcare analytics and decision-support systems using claims and provider data to support financing reform, provider performance, anomaly detection, and operational monitoring.
- Worked with datasets covering more than 9M insured individuals and 4,500+ healthcare facilities.
- Built large-scale pipelines processing 100M+ healthcare claims records using Python and SQL.
- Developed monitoring systems with anomaly detection and automated indicators across 1,000+ providers.
- Reconciled 5,000+ drug SKUs across seven agencies using text similarity and embedding-based matching.
ZF Group: pricing leakage detection and optimization
Developed anomaly detection, elasticity, uplift, and optimization models for commercial pricing decisions across a large industrial product portfolio.
- Protected approximately EUR11M annually by identifying pricing inconsistencies and channel leakage.
- Built pricing and uplift models across a EUR500M+ portfolio, contributing to a 13% gross-profit improvement.
- Deployed PySpark and Databricks pipelines with engineering partners for commercial analytics workflows.
Mobile health engagement and demand forecasting
Built forecasting, survival, and experimentation workflows for pharmacy supply chains, healthcare workers, and mHealth applications.
- Forecasted demand for pharmacy networks using DeepAR and time-series modeling, reducing stockouts by 18%.
- Improved engagement outcomes with survival modeling, churn prediction, contextual bandits, and rule-based optimization.
- Designed A/B testing, multi-armed bandit, and reinforcement-learning approaches for adaptive interventions.
AI and recommender systems
Engineering projects
Multi-Modal LLM-based Product Recommender System
Built a recommendation system that combines product reviews, metadata, images, temporal ordering, and multimodal feature fusion to predict the next items a user may purchase or review.
- Built a multimodal recommender over approximately 3M Amazon interactions using review text and image embeddings.
- Fine-tuned a GPT-2 style recommendation model with textual, visual, and temporal features.
- Implemented train/validation/test processing that respects time ordering to reduce leakage.
- Reported NDCG@5 of 0.22 and P@5 of 0.29, with evaluation harnesses, ablations, and documentation.
Deep Contextual Bandits
Adapted deep contextual bandit ideas for reusable experimentation workflows, focusing on Bayesian neural networks, Thompson sampling, and decision-making under uncertainty.
- Grounded in the ICLR 2018 Deep Bayesian Bandits benchmark.
- Built toward a package that can run contextual bandit algorithms on arbitrary datasets.
- Connects directly to product experimentation and adaptive intervention design.
Business ML: retention, attribution, and segmentation
Converted online retail transactions into a business analytics workflow covering revenue, retention, customer growth, attribution, journey analysis, and RFM-based segmentation.
Publications
Research and papers
Enhancing Product Recommendations with Multi-Modal LLMs
Research on multimodal product recommendation using text and image representations for next-item prediction.
Data-Driven Approach to Capitation Reform in Rwanda
Claims-data-driven capitation design, calibration, monitoring, and prescribing-quality insights for Rwanda's Community-Based Health Insurance scheme.
User Engagement in Mobile Health Applications
Probabilistic and survival-analysis framework for engagement and churn in mobile health applications used by healthcare workers.
Power Samade distribution: properties and application to real lifetime data
Nigerian Journal of Science and Environment paper on distributional modeling and lifetime data analysis.
Homework vs. In Class-Exercise: Means of Assessment, Waste of Time or Punishment?
International Journal of Scientific and Engineering Research.
Writing
Selected articles
Designing Recommendation Systems for Search in E-commerce
System-design oriented discussion of retrieval, ranking, and search recommendation tradeoffs.
Designing Machine Learning Solution for Course Recommendation
End-to-end framing of a course recommendation problem from business goal to ML design.
Customer spend, satisfaction, and segmentation
Marketplace analytics using customer segmentation, satisfaction prediction, and spend modeling.
Identifying networks in customer reviews
Network analysis applied to customer review relationships and behavioral insight discovery.
Selected earlier work
Analytics breadth
Is the movie industry dying?
Explored film revenue, budget, genre, rating, audience, cast, and director effects to advise production strategy.
The determinants of happiness
Applied statistical modeling to study drivers of subjective wellbeing and socioeconomic outcomes.
Is comparison really the thief of joy?
Empirical analysis using South African data to study comparison, life satisfaction, and economic context.
Data visualization practice
Built a reference collection of Python visualization patterns inspired by The Economist, data-to-viz, and storytelling-with-data practices.