16-year-old Kenyan student building enterprise-grade climate software.
I am the founder of Kenya Climate Data Lab — an open-source, county-level drought and vegetation stress monitor for all 47 Kenyan counties.
Instead of building a static dashboard, I built a full-stack intelligence service that uses satellite data, machine learning, and automated CI/CD pipelines to predict agricultural risk.
- Interactive Dashboard (Streamlit) — User-friendly interface with ML-powered next-month drought predictions.
- Developer API (FastAPI) — REST API for programmatic access to the dataset.
I designed and deployed the entire stack from scratch:
- Data Engineering: Python, Pandas, CHIRPS (rainfall), Sentinel-2 (NDVI), ERA5-Land (soil moisture). Automated data ingestion pipelines using GitHub Actions.
- Machine Learning: XGBoost classifier trained on lagged soil moisture and rainfall anomalies to predict severe drought probability (ROC-AUC: 0.82).
- Backend: FastAPI, Uvicorn, Pytest. Deployed on Render.
- Frontend: Streamlit, Plotly. Deployed on Streamlit Community Cloud.
- Infrastructure: Docker, Docker Compose, GitHub Actions (CI/CD), Run Manifests.
My research (currently being prepared for preprint) uncovered three critical insights:
- Drought ≠ Crop Damage: In 2022, 31 counties experienced severe drought, but only 8 experienced it while maize was actively growing.
- Soil Moisture Leads Vegetation: Topsoil moisture anomalies predict vegetation collapse by 1 month (r = 0.461).
- Complementary to NDMA: The monitor tracks crop-region vegetation stress in all 47 counties, complementing NDMA's pastoral drought tracking.
