An analytics and exploration tool for humanitarian funding and crisis severity, combining INFORM Severity, FTS funding, and CBPF allocation datasets to surface where needs and funding are most mismatched in 2025.

• 3D Globe — interactive globe with country hover/click, severity glow, and spike overlays switchable between funding-gap and severity modes. Supports dot-cloud and solid-fill map styles.
• Overview Tab — underlooked crises and absolute funding-gap rankings, underlooked countries table (neglect index), and global funding progress cards.
• Countries Tab — searchable country list with sort/filter controls (severity, funding gap, neglect index). Clicking a country focuses the globe and shows country detail.
• Country Detail — severity summary, funding gap bar, per-country crisis list, appeal breakdown, and CBPF cluster analysis.
• Crises Tab — category-based crisis browsing with per-category aggregate stats and crisis detail view.
• AI Chat Panel — embedded RAG chatbot (LangChain LCEL + Pinecone) that answers questions about humanitarian data. Responses that identify a primary country focus the globe and switch to the Countries tab automatically.

• Framework: Next.js 16 (App Router, Turbopack)
• 3D: Three.js via React Three Fiber
• Styling: Tailwind CSS + shadcn/ui
• AI: LangChain LCEL · Pinecone vector store · OpenRouter (Llama-3) · HuggingFace (Mistral-7B fallback)
• Data: INFORM Severity, FTS requirements/funding CSV, CBPF specificcrisis CSV, GeoJSON country borders

npm install
cp .env.local.example .env.local   # add OPENROUTER_API_KEY, HF_TOKEN, PINECONE_* keys
npm run dev

Open http://localhost:3000.

Populate the Pinecone vector store from the source CSVs:

npx ts-node scripts/ingest-data.ts

See AGENT.md for detailed state, data modelling, and engineering notes.

Neglect index: (severity / 5) × (1 − % funded) × log₁₀(1 + requirements / $1M)

Globe coordinate mapping puts the prime meridian facing the camera: x = cos(lat)·sin(lon), y = sin(lat), z = cos(lat)·cos(lon)