Guarded Exam is a web-based exam submission system designed to enhance academic integrity using advanced Natural Language Processing (NLP). It automatically detects AI-generated answers and grades student submissions based on semantic similarity to model answers—without intrusive monitoring.
- ✍️ Exam Form Interface – Simple, secure web form for student answers.
- 🤖 AI Detection Module – Identifies GPT-like AI-written content using fine-tuned transformer models.
- 🧠 Similarity Scoring – Automatically grades answers based on semantic similarity to tutor-provided answers.
- 📊 Instructor Dashboard – View results, detection outcomes, and grades.
- 🐳 Dockerized Deployment – Fully containerized via Docker and Kubernetes for easy scaling and reproducibility.
- Backend: Django
- Frontend: HTML/CSS
- Models:
RoBERTa-large OpenAI Detecor,GPT-2(AI Detection)DeBERTa-v3-zero-shot,DeBERTa Cross-Encoder(Similarity Scoring)
- Database: SQLite
- Deployment: Docker + KinD (Kubernetes in Docker)
# Clone the repository
git clone https://github.com/Mexbow/Guarded-Exam.git
cd guarded-exam
# Build and start containers
docker-compose up --buildOr deploy via Kubernetes:
# Load Docker images into KinD
kind load docker-image guarded-exam-api
# Apply Kubernetes manifests
kubectl apply -f k8s/| Module | Metric | Score |
|---|---|---|
| AI Detection | Accuracy | 98.8% |
| Similarity Model | Accuracy | 93.0% |
| False Positives | AI Detection | < 3% |
| False Positives | Similarity Model | < 10% |
All models were fine-tuned on large-scale datasets such as AI vs Human Text, Quora Question Pairs, and SciTLDR.
- AI vs Human Text (Kaggle)
- DAIGT-v2 Dataset
- Human vs LLM Text Corpus
- Quora Question Pairs
- STSB Multi-MT
- AllenAI SciTLDR
This project is licensed under the MIT License. See the LICENSE file for more details.
- Hugging Face Transformers
- Django REST Framework
- Docker & Kubernetes community
- Our supervisors at FCAI-HU
- Front Page:
- Admin View:
- Teacher View:
- Student View:
