Business intelligence and strategic analytics — KPI frameworks, dashboards and AI-augmented analytics for people who have to decide.
Six years at the Global Fund building the reporting behind multi-billion-dollar health investments; on the side, CFO and analytics lead of Blueplates, an Amsterdam EV-mobility scale-up. SQL and Python by hand; increasingly, local LLMs doing the boring parts with a human in the loop.
| Project | What it is | Stack |
|---|---|---|
| Common Health System Indicators · live | One-screen country profile: 12 health-system indicators for every WHO member state, drawn against income-group peers, 2010 → latest. Data refreshed from WHO, World Bank, UNICEF and UN MMEIG APIs into Parquet. | Python · Streamlit · Plotly |
| USG MoU dashboard · live | Tracks the 34 America First Global Health Strategy bilateral MoUs (2026–2030): US funding taper vs government co-financing, programmatic targets, full line-item data. Tables machine-transcribed from the source PDFs and verified. | Python · Streamlit · Plotly · standalone HTML |
| Budget Query Assistant | Ask a budget database a question in plain language: local model proposes SQL, a guard decides, a read-only database answers — or it asks a clarifying question, or declines. Sanitised public rebuild of an assistant in daily use. | Python · Streamlit · sqlglot · Ollama |
- Governed core, fast edge. The BI platform stays the single source of truth; around it, Python/Plotly tools and LLM assistants iterate in hours instead of weeks.
- Answer or decline, never guess. Natural-language-to-SQL assistants that ask a clarifying question or say cannot answer rather than inventing a number.
- Local by default. Where data is confidential, the model runs on-premises.
- Quality loops, not reports. Data-quality checks that route back to the source owner so the fix lands upstream.
SQL Server · Python (pandas, Plotly, Streamlit, openpyxl) · Azure (Data Factory) · Tableau · Power BI · local LLMs and LLM APIs · Claude Code
LinkedIn · Geneva / Annecy