Live BLS macro readings · AI-exposure forecasts across 342 U.S. occupations · Real-time AI/job-loss news
KAR (Karpathy AI Researcher) is an open-source, full-stack intelligence platform that maps the U.S. labor market through the lens of AI disruption. Inspired by karpathy.ai/jobs by Andrej Karpathy, KAR connects live Bureau of Labor Statistics data with LLM-powered analysis to score every major U.S. occupation across six AI-exposure dimensions — giving workers, researchers, and policy-makers a real-time window into the future of work.
🔭 Covering ~143M U.S. jobs across 342 occupations — scored, mapped, and forecasted with AI.
| Feature | Description |
|---|---|
| 📊 Live BLS Macro Dashboard | Real-time unemployment, labor-force participation, nonfarm payrolls & earnings, refreshed every 2 hours via the BLS public API |
| 🤖 AI Exposure Scoring | 342 occupations scored across 6 dimensions: cognitive load, digital dexterity, physical-world barrier, creative originality, social-emotional bandwidth, economic substitutability |
| 🔭 Horizon Forecasting | Employment, wage, and AI-exposure trajectories across 8 time horizons (30d to 5y) |
| 🗺️ Labor Market Map | Interactive grid visualization of occupations plotted by AI exposure vs. growth rate |
| 💬 Ask KAR | Conversational AI agent for querying any labor market data, forecasts, or occupation details |
| 📁 Workforce Upload | Upload and analyze custom workforce files for organizational planning and risk assessment |
| 🧠 Software 3.0 Analysis | Karpathy's Software 3.0 framework applied to labor market predictions |
| 🔔 Anomaly Alerts | Automatic detection of unusual labor market signals with LLM-generated narratives |
| Layer | Technology |
|---|---|
| Framework | TanStack Start v1 — full-stack React 19 with SSR/SSG |
| Build Tool | Vite 7 |
| UI | React 19, Tailwind CSS v4, Radix UI |
| Charts | Recharts |
| Backend | Supabase — database, auth, storage, realtime |
| Server Functions | TanStack createServerFn (edge runtime) |
| Data Fetching | TanStack Query v5 |
| Type Safety | TypeScript 5.8 + Zod |
| Package Manager | Bun |
# 1. Clone the repo
git clone https://github.com/eli-devop/KAR.git
cd KAR
# 2. Install dependencies
bun install
# 3. Copy environment variables
cp .env.example .env
# Edit .env with your Supabase credentials
# 4. Start the dev server
bun devOpen http://localhost:3000 — you should see the live dashboard.
bun run build
bun run startKAR/
├── src/
│ ├── routes/ # File-based routing (TanStack Router)
│ │ ├── index.tsx # Dashboard — live BLS macro tiles, KPIs, news
│ │ ├── map.tsx # Labor Market Map (occupation grid)
│ │ ├── chat.tsx # Ask KAR (conversational AI agent)
│ │ ├── capabilities.tsx # AI capability dimensions explorer
│ │ ├── scoring.tsx # Scoring Studio
│ │ ├── workforce.tsx # Workforce file upload & analysis
│ │ ├── reports.tsx # Reports
│ │ └── __root.tsx # Root layout (sidebar, nav, providers)
│ ├── components/
│ │ ├── kar/ # KAR-specific components
│ │ └── ui/ # shadcn/ui base components
│ ├── lib/ # Business logic & server functions
│ │ ├── forecast.functions.ts
│ │ ├── news.functions.ts
│ │ └── alerts.functions.ts
│ ├── hooks/ # Custom React hooks
│ ├── contexts/ # React contexts (HorizonContext)
│ ├── data/ # Static data (occupations, BLS series IDs)
│ └── styles.css # Tailwind v4 entry + design tokens
├── supabase/ # Database migrations
├── public/ # Static assets
└── html/ # Static HTML reference pages
Create a .env file in the project root (see .env.example):
VITE_SUPABASE_URL=https://your-project.supabase.co
VITE_SUPABASE_PUBLISHABLE_KEY=your-anon-key
SUPABASE_SERVICE_ROLE_KEY=your-service-role-keyEvery 2 hours, KAR automatically:
- Fetches the latest BLS macro series (unemployment, payrolls, participation, earnings)
- Scrapes AI and job-loss news from trusted sources
- Detects labor market anomalies and generates alerts
- Updates forecast narratives using LLM analysis
You can also trigger a manual refresh from the dashboard or via the public API endpoint.
Each occupation is scored on a 0–100 scale across six dimensions:
| Dimension | What It Measures |
|---|---|
| 🧠 Cognitive Load | Abstract reasoning, problem-solving complexity |
| 💻 Digital Dexterity | Existing software/tech usage in the role |
| 🏗️ Physical-World Barrier | Degree of physical manipulation required |
| 🎨 Creative Originality | Novel ideation vs. pattern-based execution |
| 🤝 Social-Emotional Bandwidth | Empathy, leadership, human connection required |
| 💰 Economic Substitutability | Cost-benefit of automating vs. hiring |
KAR projects across 8 time horizons to support different planning cycles:
| Horizon | Use Case |
|---|---|
| Now | Current BLS readings + live news |
| 30d | Near-term policy shifts, earnings season |
| 60d | Quarterly rebalancing signals |
| 90d | Budget-cycle impacts |
| 1y | Annual planning, election-year policy |
| 2y | Mid-cycle technology adoption |
| 3y | Workforce restructuring, Fed targets |
| 4y | Presidential term horizon |
| 5y | Long-range strategic planning |
KAR is built on Andrej Karpathy's Software 3.0 vision — where natural-language prompts are the program. The platform shows how Software 1.0 (deterministic code), Software 2.0 (trained ML models), and Software 3.0 (LLM-powered reasoning) combine to produce actionable labor market intelligence.
- OAuth login — save personal watchlists and custom forecasts
- Occupation comparison — side-by-side AI-exposure analysis
- CSV/PDF export — download full scoring reports
- API v1 — public REST endpoints for researchers
- Employer mode — bulk workforce analysis for HR teams
- Webhook alerts — push notifications on BLS data releases
- i18n support — multilingual labor market data
Contributions, issues, and feature requests are welcome! See CONTRIBUTING.md for guidelines.
# Fork the repo, create a feature branch, then:
bun install
bun dev
# Before opening a PR:
bun run lint && bun run format
bunx tsc --noEmitPlease read our Code of Conduct before contributing.
Found a vulnerability? Please do not open a public issue. Email the maintainer directly at dmesa@e-l-i.net. We take security seriously and will respond within 48 hours.
- Andrej Karpathy — for karpathy.ai/jobs, the original inspiration and Software 3.0 framework
- U.S. Bureau of Labor Statistics — for the Occupational Outlook Handbook data
- Supabase — for the open-source backend platform
- TanStack — for the incredible full-stack React toolkit
If KAR helps you understand the future of work, please consider giving it a ⭐ — it helps more people find the project!