Turning large, messy, multi-source datasets into insights and decisions for stakeholders.
Hello there, my name is Stephane Ngugi and I am currently working as a data analyst and researcher with 5+ years across civic-tech, start-ups, and non-profit data teams, working mostly in Python, R, and SQL. I have also worked as a data scientist in one of my former roles. My work lives where large-scale dataset specifically benchmarking, big-data analytics, and building dashboards; I spend a lot of my time on data mining and extraction, data cleaning, data visualizations, web scraping, API integrations and Data pipelines.
- 🔭 Currently building: agentic data workflows and large-scale big data pipelines, and I am working toward AI analytics engineering.
- 🤖 Side projects: washamba bots, my team's agentic system.
- 📚 Learning: AI safety and governance through the BlueDot Impact community and a monthly AI Governance Club meetings.
- 🏆 In public: Kaggle competitions and I help host Hackathons and conversations in Kenya's Claude developer community.
Languages
Data Science & Machine Learning
Big Data, Databases & Cloud
BI, Dashboards & Visualisation
Web Scraping & Data Extraction
Geospatial & Remote Sensing
Workflow & Tooling
Methods
- Satellite field-boundary detection: benchmarked five deep-learning segmentation architectures (U-Net+ResNet34, DeepLabV3+, PSPNet, FPN, Inception) on multi-temporal Sentinel-2 imagery for smallholder plots in Rwanda; presented at the Kaggle workshop, ICLR 2023.
- Satellite CO₂ emissions regression: predicted GRACED-derived CO₂ across 105 Rwandan locations from Sentinel-5P co-emitted pollutant observations, a replicable approach for climate-finance MRV where ground monitoring is absent.
- NLP disclosure extraction: Python pipeline using Amazon Textract and OpenAI embeddings with cosine similarity to pull geographic-footprint data from 2,000+ corporate disclosures, replacing a manual research workflow.
- Fraud-detection benchmarking: reconciled PostgreSQL and MS SQL Server sources in Python, computed Value Detection Rate and False Positive Rate across 15+ financial institutions, and served the results through a Power BI and Streamlit self-service layer.
Studying AI safety & governance · forecasting · open data for public good