JUD:GE - Just-in-time Unified Deployment: Grading Engine

JUD:GE is an inference-only online judge for Python HTTP servers. Users submit a ZIP that runs python app.py, and the platform evaluates accuracy/latency on hidden test cases.

Highlights

• Node.js (ESM) backend with Express
• BullMQ + Redis async queue
• Docker-based grading (resource limits, read-only rootfs)
• React + TypeScript frontend (front/)
• Prisma ORM (SQLite for dev, PostgreSQL-ready)

Repository Structure

/server.js
/worker.js
/package.json
/prisma/schema.prisma
/src/...
/uploads (gitignored)
/data (gitignored)
/docs/...
/front/...

Prerequisites

• Node.js 18+
• Docker Engine
• Redis
• Python 3.11+ (for evaluator scripts)

Setup (Development)

1. Install dependencies:

npm install
npm --prefix front install

2. Create env file:

copy .env.example .env

3. Run Prisma migrations:

npm run prisma:migrate

4. Seed demo data:

npm run seed

Default Admin Account

After seeding, a default admin account is created:

• email: admin@judge.local
• password: admin123

5. Start dev servers:

npm run dev

Setup (Production)

1. Build frontend:

npm --prefix front run build

2. Start API server:

npm start

3. Start worker:

npm run worker

Key Endpoints

• POST /api/auth/register
• POST /api/auth/login
• GET /api/problems
• POST /api/submissions (multipart zip upload)
• GET /api/leaderboard/:problemId
• POST /api/admin/settings

Submission Format

See docs/SUBMISSION_SPEC.md for detailed instructions.

Notes

• The backend serves front/dist when it exists.
• uploads/ stores artifacts and zips.
• data/ stores datasets, evaluators, and run logs.