CrisisLens

CrisisLens is a monorepo with a Next.js command-center UI and a Python ML workspace.

Repository Layout

• apps/web: Next.js 14 app (App Router), React 18, TypeScript, Three.js globe, API routes, and tests.
• apps/ml: Python model training code and model artifacts.

Prerequisites

• Node.js 20+
• pnpm (workspace package manager)
• Python 3.10+

Web App (apps/web)

Install

pnpm install

Run locally

pnpm run generate:data
pnpm run dev

Open http://localhost:3000.

Common commands

pnpm run lint
pnpm run typecheck
pnpm run test
pnpm run test:unit
pnpm run test:e2e
pnpm run test:all

All root scripts delegate to apps/web.

ML Workspace (apps/ml)

Setup

python -m venv .venv
source .venv/bin/activate
pip install -r apps/ml/requirements.txt

Train (current script entry)

python apps/ml/models/train_model.py

Notes

• Frontend tests live in apps/web/tests (Vitest + Playwright).
• Web data generation script lives at apps/web/scripts/generate-country-metrics.mjs.
• Optional env var: NEXT_PUBLIC_GLOBE_WS_URL for real-time globe highlights.

Geo-Insight Agent Endpoints

Set these env vars in apps/web/.env.local:

• DATABRICKS_HOST
• DATABRICKS_TOKEN
• DATABRICKS_WAREHOUSE_ID
• CRISIS_TABLE_FQN
• AI_MODEL

Optional compatibility overrides (if your workspace does not expose AI Gateway path):

• DATABRICKS_AI_CHAT_PATH (example: /api/2.0/ai-gateway/chat/completions)
• DATABRICKS_AI_ENDPOINT (serving endpoint name to call via /serving-endpoints/{name}/invocations)

Quick checks:

curl -s "http://localhost:3000/api/geo/metrics?iso3=HTI"

curl -s "http://localhost:3000/api/geo/insight?iso3=HTI"

curl -s -X POST "http://localhost:3000/api/geo/query" \
  -H "Content-Type: application/json" \
  -d '{"question":"Which countries have lower funding coverage than Mali and where should funding increase?"}'