Globe: a geopolitical intelligence dashboard with an AI analyst that drives the map

Explore any country

Click a country (or search for one) and the globe flies to it and highlights it. The panel opens on an Overview with the country's profile and headline numbers. The KPIs tab splits the 22 indicators into Economic, Fiscal, Governance, Trade, People and Infrastructure. Each card shows the latest value, the change and a 5-year chart.

Paint the planet

The Heatmap button colors every country by any of the 22 indicators, with a legend in the corner. It works on both the 3D globe and the flat 2D map. Plain values such as electricity access or GDP use a light-to-dark blue scale. Values that cross zero, like governance scores, switch to a red–white–blue scale. A design note says it deliberately avoids red/green "good/bad" judgments on neutral data. Countries with no data show in gray.

Head to head

"Compare with another country" puts two countries side by side. The Compare tab shows proportional bars for the headline numbers. Compare KPIs pairs their 5-year histories indicator by indicator.

The AI analyst

Everything above, the AI can do too. Every click, search and hover goes through a single dispatch() command system, and the AI was designed from day one to send the same commands. Answering is split into two phases:

• Research: the agent runs several web searches from different angles, pulls country data, gets the countries in a region and ranks countries by an indicator. It is told explicitly not to write an analysis yet. Each step shows up live in the panel's Agent tab.
• Respond: with everything gathered, it streams its answer and uses its visual tools. show_choropleth paints the globe and annotate_map pins colored, scored labels on countries. All the scores in one call must be on the same scale.

The annotation layer: Southeast Asia labelled with 2024 GDP growth (%), sent through the same command the AI uses

The Intelligence Assistant panel with its suggested questions

Where it stands

The project runs locally as a React frontend (port 5173) and a FastAPI backend (port 8000). Everything in the screenshots works without any API keys. The AI assistant needs a DeepSeek key. News is still static sample data for 10 countries. The original plan (frontend/og plan.txt) started as a frontend with mocked data only, and the project has grown well past that.

Open questions

• Is there a target audience in mind (analysts, investors, students)? It would shape what comes next, e.g. live news.