py2cpp — AI Python → C++ (CF-ready)

py2cpp converts Python to C++17 with a Codeforces-style boilerplate, then compiles and runs both versions side-by-side on the same input.
AI-first: Qwen2.5-Coder (via Ollama) + a compile-and-repair loop.
Live UI: https://py2cpp.vercel.app/

---

Requirements

• macOS with Homebrew
• Python 3.10+, Node 18+
• Ollama installed

---

Quick start

1. Model server

brew install ollama cloudflare/cloudflare/cloudflared
ollama serve
ollama pull qwen2.5-coder:7b

2. Backend

cd backend
python3 -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
export OLLAMA_MODEL=qwen2.5-coder:7b
uvicorn main:app --host 0.0.0.0 --port 8000

3. Frontend

cd frontend
echo "VITE_API_URL=https://<your-tunnel>.trycloudflare.com" > .env
npm install && npm run dev

The frontend is static. The backend (FastAPI + Ollama + g++) runs on your machine and must stay running.

4. Public URL

cloudflared tunnel --url http://localhost:8000
# copy the https://<random>.trycloudflare.com URL

Notes: Deployment can be only for frontend via netlify or vercel ; backend would be running locally constantly to work

API

• POST /ai/convert → body { "py": "" } → { "cpp": "<full c++ file>" }

• POST /run/python → { "code": "", "stdin": "" }

• POST /run/cpp → { "code": "<c++>", "stdin": "" }

• GET /healthz → { "ok": true }

Key points:

• macOS/Clang is handled by a fallback header set when <bits/stdc++.h> is unavailable.

• Targeted at contest-style Python; highly dynamic features may require manual edits.

License

MIT © Birajit Saikia