Epidemiologist & Data Scientist · Real-World Evidence · Causal Inference · Healthcare AI & Analytics
📍 Richmond, Kentucky, USA | 🇺🇸 U.S. Permanent Resident — No Sponsorship Required
I have spent 17+ years in epidemiology, global health analytics, data science and agentic AI, underpinned by doctoral training in epidemiology. I build the analysis and the system around it: cohort and survival studies on real-world data, causal designs for policy questions, and the pipelines, models and apps that put results in front of decision makers.
Core areas (each links to a repo that shows it):
- 🔬 Real-world evidence & causal inference: oncology cohort and survival analysis, IPTW and immortal-time-bias methods, difference-in-differences and event-study designs
- 📊 Health economics: Medicaid coverage and budget impact
- 🧠 Machine learning for health: risk stratification with SHAP and calibration
- 🤖 AI / LLM systems: production RAG, multi-agent orchestration
- 🏗️ Data platforms: Microsoft Fabric, Databricks lakehouse
| Project | What it shows | Stack |
|---|---|---|
| oncology-rwe-nsclc · report | NSCLC survival pipeline on synthetic EHR data (Synthea → OMOP CDM → dbt → R), benchmarked against 172,582 U.S. patients from SEER. Documents a case where the synthetic data failed validation: unmodified Synthea staged every lung cancer as stage I. | R · OMOP CDM · dbt · PostgreSQL |
| nsclc-rwd-truth-recovery | Companion project: a custom generator with a known truth, rule-based line-of-therapy derivation (98.1% whole-sequence exact match against truth), IPTW, an immortal-time-bias demonstration, and a 10-seed Monte-Carlo bias and coverage study. | Python · lifelines · scikit-learn |
| incretin-access-value · site | Public-data study of Medicaid coverage of Wegovy and Zepbound: 12.2 additional prescriptions per 1,000 enrollees per quarter (95% CI 8.4 to 16.0) and a net budget impact of about $161.3 million over five years for a 1-million-enrollee program. Tested warehouse, Quarto report, interactive budget model. | PostgreSQL · dbt · R · Quarto · Shiny |
| Project | Description | Stack |
|---|---|---|
Immunization Defaulter Risk Engine  |
XGBoost pipeline predicting vaccine defaulter risk for 6,864 children across 4,672 CHW areas in Kenya. Per-patient SHAP explainability, isotonic calibration (ECE 0.023), PSI drift monitoring, FastAPI serving, Streamlit dashboard. | Python · XGBoost · SHAP · FastAPI · PostgreSQL · MLflow · Optuna |
| Databricks Medicare Lakehouse | Medicare Advantage lakehouse: medallion ETL, CMS HCC v28 risk adjustment, XGBoost + SHAP, Unity Catalog governance. | Databricks · Delta Lake · Python |
| Community Health Intelligence Platform | Community health lakehouse for Kenya's CHW program: medallion pipeline, Unity Catalog row-level security, AI/BI Genie, dashboard. | Databricks · Delta Lake · dbt · Airflow |
| Insurance Premium Prediction | End-to-end ML pipeline with CI/CD, MLflow tracking and SHAP explainability. | Python · XGBoost · MLflow · SageMaker |
| Within Reach | #TidyTuesday analysis of hospital access across 11,422 U.S. cities, with a live Shiny app. | R · Shiny |
| Project | Description | Stack |
|---|---|---|
| Clinical Document Intelligence | FDA drug-label RAG with 5-stage retrieval, multi-agent orchestration, clinical guardrails and 54 automated tests. | Python · FastAPI · ChromaDB |
| AgentCare | Six-agent LangGraph orchestration for patient administration and care coordination. | Python · LangGraph · FastAPI · Next.js |
| Women's Health RAG | RAG over Demographic and Health Survey reports. | Python · LangChain · LangGraph · pgvector |
| AI-Powered Research Assistant | RAG platform for scientific papers with modular LangGraph workflows. | Python · LangGraph · Pinecone |
| Healthcare Q&A RAG Platform | Healthcare knowledge retrieval with vector search and role-based access. | Python · FastAPI · ChromaDB |
| Multimodal PDF RAG System | Document intelligence with OCR, table extraction and semantic search. | Python · FastAPI · React |
| Project | Description | Stack |
|---|---|---|
| CertiAce Retail Analytics | Fabric portfolio project covering the DP-600 domains: medallion (Bronze/Silver/Gold) architecture over 555K rows, semantic model with DAX measures, KQL real-time monitoring, RLS/OLS and deployment pipelines. | Microsoft Fabric · OneLake · KQL · DAX · Power BI |
| E-Commerce Intelligence Platform | Lakehouse pipeline: 1.7M rows, 6 relational tables, medallion architecture, conversion and segmentation models tracked in MLflow. | Databricks · Delta Lake · MLflow |
| What I built | Result |
|---|---|
| Immunization defaulter risk engine (Kenya MOH eCHIS) | ROC-AUC 0.892 · 6,864 children · 4,672 CHW areas · live app |
| Medicaid obesity-drug coverage study | +12.2 prescriptions per 1,000 enrollees per quarter (95% CI 8.4 to 16.0); about $161.3M net over five years per 1M enrollees · incretin-access-value |
| Oncology RWD pipeline validated against a known truth | 98.1% line-of-therapy exact match; 10-seed Monte-Carlo bias and coverage · nsclc-rwd-truth-recovery |
| CertiAce Retail Analytics (Fabric portfolio) | Medallion lakehouse · 555K rows · DAX · KQL · RLS/OLS · deployment pipelines |
| ML predictive models for health outcomes | 30% improvement in prediction accuracy |
| Automated data pipelines (AWS + PostgreSQL + dbt) | Reporting turnaround: 10–14 days → near real-time; manual processing down 40% |
| Causal inference and RWE studies | 25+ studies (PSM, IPW, DiD, ITS, TMLE) informing $2M+ in annual resource allocation |
| Data and analytics for community health programs | Programs reaching 8.5M+ people in Kenya, Uganda and Burkina Faso |
| Peer-reviewed publications | 30+, including The Lancet (2025) |
| Skill | Evidence |
|---|---|
| Survival analysis (Kaplan–Meier, Cox, time-varying exposure) | oncology-rwe-nsclc · nsclc-rwd-truth-recovery |
| Pharmacoepidemiology methods (IPTW, immortal-time bias, E-value) | nsclc-rwd-truth-recovery |
| Causal inference (difference-in-differences, event study) | incretin-access-value |
| OMOP CDM, dbt, PostgreSQL | oncology-rwe-nsclc · incretin-access-value |
| Budget impact / health economics | incretin-access-value |
| R, Quarto, Shiny | incretin-access-value · within-reach |
| Python, scikit-learn, lifelines | nsclc-rwd-truth-recovery |
| XGBoost, SHAP, calibration, drift monitoring | immunization-defaulter-risk-engine |
| MLflow, Optuna | immunization-defaulter-risk-engine · insurance-premium-prediction-ml |
| FastAPI, Docker, Streamlit | immunization-defaulter-risk-engine |
| AWS SageMaker | insurance-premium-prediction-ml |
| PyTorch (CNN) | FreshHarvest |
| RAG, vector search (ChromaDB, pgvector, Pinecone) | clinical-doc-intelligence · womens-health-rag · AI-Powered-Research-Assistant |
| LangChain, LangGraph, multi-agent systems | AgentCare · clinical-doc-intelligence |
| Microsoft Fabric, OneLake, KQL, deployment pipelines | certace-fabric-analytics |
| Power BI, DAX, RLS/OLS | certace-fabric-analytics |
| Databricks, Delta Lake, Unity Catalog | community-health-intelligence-platform · databricks-medicare-lakehouse |
| Airflow | community-health-intelligence-platform |
| Degree | Institution |
|---|---|
| PhD, Epidemiology | Jomo Kenyatta University of Agriculture and Technology (JKUAT) |
| MSc, Health Systems Management | Kenya Methodist University |
| BSc, Statistics | University of Nairobi |
| Certification | Issuer | Status |
|---|---|---|
| Microsoft Certified: Power BI Data Analyst Associate (PL-300) | Microsoft | ✅ |
| Google Data Analytics Professional Certificate | ✅ | |
| CITI Program: Human Subjects Research; Good Clinical Practice; Big Data and Data Science Research Ethics | CITI Program | ✅ |
| Microsoft Fabric Analytics Engineer (DP-600) | Microsoft | In preparation |
I pair statistical rigor with working software. I check my methods against a known answer where I can, document where a data source fails validation, and ship analyses as reproducible pipelines. The NSCLC truth-recovery study and the Synthea-versus-SEER benchmark are examples.
Hands-on and leadership roles across real-world evidence, epidemiology, health data science and applied AI:
- Real-World Evidence Scientist / Epidemiologist
- Senior / Principal / Lead Data Scientist (healthcare and clinical)
- Population Health Analytics Lead
- Analytics Engineer / Microsoft Fabric Engineer
- Director / VP, Data & Analytics
Target sectors: Pharma · Biotech · CRO · Health Systems · Payers & Insurers · Health Tech · Global Health · Federal Contractors
📩 keyegon@gmail.com | 🔗 LinkedIn | 🆔 ORCID