All the tables.
One clear story.

What does maize cost today — in Harare, or in any of the hundreds of markets WFP publishes across Africa? The data exists, scattered across WFP, national agencies, and a dozen formats: raw CSVs and PDFs, no unified API. SokoData turns that scatter into one clean, documented, queryable API — African market prices at its core, plus a deep Zimbabwe data commons (economy, climate, demographics). 22 datasets, all open data, provenance per record.

Live right now: 486 markets, 31 commodities, daily USD price movers, CPI, exchange rates, fuel, and climate series — one ETL per dataset, one catalog, and analytics hardened by the world's messiest currency history.

An open database documenting who finances, designs, builds, and operates African infrastructure — across any period, including pre-2000 and archived sources. Neutral by design: if evidence only shows financing or equipment supply, the record says exactly that. Every material claim links to a source, a verbatim passage, and an evidence-strength label.

The product is the database, not the map: JSON Schema validation with pydantic and Frictionless, PR-reviewed curated data, and versioned releases in CSV / GeoJSON / GeoPackage / SQLite. The interactive map (Astro, Svelte, MapLibre) is just the exploration layer.

End-to-end pipeline: EDA → feature engineering → gradient boosting. The model finds at-risk customers, but the EDA finds why — tenure, balance, and activity patterns that a branch manager can act on tomorrow.

Analyzed 1,000 records to find who actually buys. High-income, middle-aged professionals converted at ~59% in the Pacific — not the largest segment, but the most valuable. The dashboard answers a manager's question in 10 seconds, not 10 minutes.

U.S. airline and airport performance across delays, cancellations, on-time rates. The stakeholder question was simple: "Which airline should we rebook on?" The answer needed one filter, not five.

2022–2023 retailer invoices: revenue trends, brand profitability, regional distribution. The story wasn't "sales went up" — it was which retailer type moved which pack size where.

Prototyped KPIs in pandas, implemented in Excel, delivered in Power BI. 2020 international performance — when "international" mostly meant "how did we cope with 2020?"