AutoGrader is a full-stack application developed at the Vibe Hack Hackathon to automate grading of Jupyter Notebooks using LLMs. Upload a .zip of .ipynb files and a .txt rubric, and receive structured feedback and scores — instantly!

Elena Nurullina Aryan Jain lenzlaww

• 📁 Upload multiple student notebooks in .zip format
• 📋 Upload a grading rubric in .txt format
• 💬 Uses Claude 3.5 Haiku via Rilla API for scoring and feedback
• 📊 Generates downloadable grading report as .csv and detailed feedback .txt
• 🖥️ Frontend built in React + Axios for smooth UX
• ⚡ FastAPI backend with async inference calls

backend/
├── app.py                   # FastAPI server http://127.0.0.1:8000/docs#/default/get_csv_csv_get
├── graders/
│   ├── ipynb_parser.py      # Notebook parser
│   ├── rubric_processor.py  # Rubric parsing logic
│   └── grader.py            # LLM interaction and grading logic
├── test_data/               # Sample submissions
├── templates/               # (Optional) For HTML rendering
├── static/                  # Static frontend files
└── requirements.txt         # Python dependencies

src/
├── components/
│   ├── FileUpload.js        # Upload .zip and rubric
│   ├── ResultDisplay.js     # Show results
├── services/
│   └── api.js               # Axios for API requests
├── App.js                   # Main React app
├── index.js                 # Entry point
└── styles.css               # Global styles

1. Upload: Send student notebooks + rubric via form.
2. Parse: Extracts code cells from .ipynb files.
3. Score: Sends rubric + code to LLM (Claude 3.5 Haiku).
4. Output: Collects scores + feedback, saves as CSV & TXT.

• Frontend: React, Axios
• Backend: FastAPI, httpx, nbformat, pandas, matplotlib
• LLM API: Claude 3.5 via LiteLLM Proxy (Rilla)
• Deployment: (Locally hosted / ready for Dockerization)

CSV:

notebook,score
student1.ipynb,88
student2.ipynb,73

Feedback (TXT):

===== student1.ipynb =====
Final Score: 88
What is wrong:
- Did not modularize functions

What can be improved:
- Use better variable names

cd backend
pip install -r requirements.txt
pip install flask fastapi uvicorn nbformat zipfile openai

uvicorn app:app --reload

cd src
npm install
npm start

• 🔒 Authentication for teacher/student roles
• 📈 Visual plots per student performance
• 🧠 Auto-rubric generation from solution notebook
• 🧪 Support for multiple LLM providers

This project was developed in under 2 hours at the Vibe Hack Hackathon, blending GenAI and education for impact.