Paper2PyTorch

Live demo link: https://paper2-py-torch.vercel.app/

Drop a research paper. Get a fully executable Jupyter notebook.

Paper2PyTorch uses four Google ADK agents arranged in a sequential pipeline to transform any ML/AI research paper PDF into runnable PyTorch code.

How It Works

PDF or arXiv URL
       │
       ▼
┌──────────────────┐
│ 01 Analyze Paper │  Extract algorithms, equations, architecture → JSON
└────────┬─────────┘
         │
         ▼
┌──────────────────────────┐
│ 02 Design Implementation │  Plan CPU-scale toy model → JSON
└────────┬─────────────────┘
         │
         ▼
┌─────────────────────┐
│ 03 Author Notebook  │  Write 12-section Jupyter notebook with real PyTorch
└────────┬────────────┘
         │
         ▼
┌───────────────────────┐
│ 04 Review & Repair    │  Fix imports, undefined vars, shape mismatches
└───────────────────────┘
         │
         ▼
  .ipynb  (download or open in Colab)

Stack

Layer	Technology
Agents	Google ADK (SequentialAgent)
LLM	Gemini 2.5 Pro
Backend	FastAPI + Python
Notebooks	nbformat
Frontend	Next.js 14 + TypeScript + Tailwind CSS
Streaming	Server-Sent Events (SSE)

Quick Start

Backend

cd backend
python -m venv venv && source venv/bin/activate
pip install -r requirements.txt

cp .env.example .env          # optionally set GOOGLE_API_KEY
python app.py                 # http://localhost:8000

Frontend

cd frontend
npm install

cp .env.local.example .env.local   # set NEXT_PUBLIC_API_URL=http://localhost:8000
npm run dev                         # http://localhost:3000

Users provide their own Gemini API key in the UI — no server-side key required.

Environment Variables

Variable	Where	Description
GOOGLE_API_KEY	Backend	Optional server-level Gemini key
GITHUB_TOKEN	Backend	Required for "Open in Colab" Gist creation
MAX_UPLOAD_MB	Backend	PDF size limit (default 30)
NEXT_PUBLIC_API_URL	Frontend	Backend URL

License

MIT