Building data infrastructure and trustworthy AI for African health systems. I build reproducible health-data pipelines, epidemiological analytics, and evaluated AI applications that turn complex health information into actionable insights.
End-to-end ETL pipeline extracting, validating, transforming, and loading Kenya county health data. Orchestrated with Apache Airflow. Includes an interactive dashboard with county risk classification and disease burden scores.
App may take 30s to wake up on first load
Deployed RAG prototype that answers natural language questions about Kenya health indicators and disease surveillance data. Every answer is grounded in source documents with citations and relevance scores via GPT-4o and Pinecone.
App may take 30s to wake up on first load
Disease surveillance dashboard combining CUSUM anomaly detection, SEIR compartmental modelling, and Prophet and XGBoost forecasting. Built for Kenya MOH contexts with early outbreak warning and scenario planning.
App may take 30s to wake up on first load
Multi-agent LangGraph system that turns a research question into a complete epidemiological study protocol including study design, sample size, confounders, and a grant-ready protocol document.
App may take 30s to wake up on first load
XGBoost clinical prediction prototype for severe malaria and TB treatment failure risk. Includes SHAP explainability for individual predictions and a demographic fairness audit identifying performance gaps across sex and age subgroups.
App may take 30s to wake up on first load
Predicts 30-day hospital readmission risk for chronic disease patients using structured clinical features with SHAP explainability to identify the patient-level factors driving each prediction.
App may take 30s to wake up on first load
ResNet18 transfer learning prototype classifying chest X-rays as Normal, Pneumonia, or TB. Generates Grad-CAM heatmaps showing which lung regions drove the prediction, making the model auditable and reviewable.
App may take 30s to wake up on first load
Production-oriented monitoring dashboard for the Kenya Health RAG system. Runs automated evaluation with GPT-4o-mini as judge, tracks faithfulness and relevancy, enforces deployment quality gates, and logs all runs to MLflow.
App may take 30s to wake up on first load
Monitoring prototype detecting data drift in the Kenya malaria prediction model across seasonal shifts, bed net distribution campaigns, and outbreak events using Evidently AI with interactive distribution comparisons.
App may take 30s to wake up on first load
Hybrid NER system combining rule-based matching and a biomedical transformer to extract diseases, study designs, populations, risk factors, locations, interventions, and statistics from East African epidemiology abstracts.
App may take 30s to wake up on first load
Four-agent LangGraph prototype for clinical decision support in East African health contexts. Agents handle retrieval from WHO protocols, differential diagnosis generation, safety validation, and clinical handover document production.
App may take 30s to wake up on first load
LLaMA-3.2 3B fine-tuned on East African epidemiology and public health literature using QLoRA on Google Colab. Trained to improve response quality on African health system contexts where base models underperform.
Live-streaming patient vitals dashboard simulating real-time data from 10 Kenya county hospitals. Auto-detects critical conditions including tachycardia, hypertensive crisis, and low SpO2 with a county-level alert feed.
Multi-tab Streamlit dashboard covering disease burden, epidemic trends, county comparison, and health system capacity across all 47 Kenyan counties (2015 to 2024). Includes a one-click PDF report generated with ReportLab covering KPIs, disease rankings, trend charts, and health system capacity gaps.
App may take 30s to wake up on first load
Multi-tab Streamlit dashboard comparing health spending, outcomes, and system capacity across Kenya, Nigeria, and South Africa (2010 to 2024). Pulls real data live from the World Bank API across 15 indicators, with linear interpolation for data gaps, a normalised radar chart, WHO-threshold gauges, and a one-click ReportLab PDF report.
App may take 30s to wake up on first load
Retrospective cohort study estimating time-to-virological-failure among HIV patients on antiretroviral therapy in Nairobi. Kaplan-Meier survival curves by regimen, log-rank tests, and Cox proportional hazards regression with adherence, WHO stage, and baseline CD4 as predictors.
Unmatched case-control study (400 cases / 400 controls) identifying modifiable and non-modifiable risk factors for hypertension in Kenyan adults. Multivariable logistic regression with bootstrap 95% CI, unadjusted vs adjusted OR comparison, PAF estimation, ROC/AUC, and Hosmer-Lemeshow calibration.
Ecological panel study modelling malaria incidence across all 47 Kenyan counties over six years (2017 to 2022). Poisson GLM with log-population offset, incidence rate ratios by ecological zone, overdispersion testing, and Negative Binomial comparison. Lake Victoria basin IRR 80 to 120 times higher than highlands baseline.
KDHS-style cross-sectional survey (N=5,400, 47 counties) estimating TB and anaemia prevalence in Kenyan adults. Dual logistic regression identifies independent determinants for each outcome, with co-occurrence analysis, HIV stratification, and county-level ranked prevalence charts.
16S rRNA amplicon sequencing analysis (V3-V4) profiling the oral microbiome of HIV-positive and HIV-negative Kenyan adults across three groups: HIV-negative controls, HIV-positive on ART, and ART-naive. Alpha and beta diversity analysis, differential abundance testing with Benjamini-Hochberg FDR, and Spearman correlations with CD4 count and viral load.
Interactive choropleth dashboard mapping disease burden across all 47 Kenya counties. Uses Global Moran's I and LISA cluster detection to identify spatial hotspots and cold spots for malaria, cholera, dengue, and measles.
App may take 30s to wake up on first load
A background in biochemistry, including epidemiology and biostatistics, gave me early exposure to health data problems. When I moved into data science I found the most consequential unsolved challenges were in African health systems: disease surveillance, clinical prediction, and health infrastructure that most AI tools aren't built for.
I'm a healthcare AI engineer building data infrastructure and machine learning systems for African health contexts. My work spans health data pipelines, epidemiological analytics, clinical NLP, and LLM applications, with a focus on developing reliable, reproducible systems that turn complex health data into actionable insights.
My approach: Precision in the data. Rigour in the methods. Impact in the real world.