Inspiration

What it does

How we built it

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for ExamGuard-X

ExamGuard-X: AI-Powered Examination Paper Leak Detection and Prevention

Inspiration

Examination paper leaks are a serious problem because they affect students' hard work, fairness, and trust in the education system. Traditional examination security methods often depend on manual monitoring and restricted access, which may not be sufficient to identify suspicious activities or detect leaked content quickly.

This inspired us to develop ExamGuard-X, an AI-powered examination security system designed to help prevent, detect, and investigate question paper leaks. Our goal is to combine modern software security techniques with artificial intelligence to make examination management more secure, transparent, and reliable for both online and offline examinations.

What It Does

ExamGuard-X aims to provide a centralized platform for managing examination papers securely throughout their lifecycle, from creation and approval to storage, distribution, and examination.

Key features include:

• Secure Paper Management: Role-based access control, authentication, encryption, and integrity verification.
• AI-Powered Leak Detection: OCR extracts text from suspected leaked documents, while semantic similarity techniques compare the content with authorized question papers.
• Anomaly Detection: Identifies unusual access patterns and potentially suspicious activities for further investigation.
• Digital Traceability: Maintains security event logs and supports watermarking to help investigators trace the possible source of a leak.
• Security Dashboard: Provides a centralized interface for monitoring examination papers, security events, alerts, and investigation results.

The system is intended to support security teams in identifying risks and investigating suspected leaks rather than claiming to eliminate every possible leakage scenario.

How We Built It

We designed ExamGuard-X using a modular architecture that separates the frontend, backend, database, and AI-based analysis components.

• Frontend: React, TypeScript, and Tailwind CSS for a responsive security dashboard.
• Backend: Java and Spring Boot for REST APIs, authentication, authorization, and examination paper management.
• Database and Caching: MySQL for persistent records and Redis for planned caching and supporting security workflows.
• AI and Document Processing: Python, FastAPI, PaddleOCR, and OpenCV for document processing and text extraction.
• Semantic Analysis: Sentence Transformers and FAISS for comparing document content and identifying potentially similar question text.
• Anomaly Detection: Scikit-learn for exploring suspicious access patterns.
• Infrastructure: Docker for containerization, with Kafka and AWS planned for event processing and deployment.

The intended workflow begins when an authorized user uploads or manages an examination paper. The backend applies access controls and security checks, while the AI pipeline processes suspected documents and compares their extracted content with authorized papers. Potential matches or unusual activities can then be presented as alerts for human review.

We are developing the project incrementally, starting with the core backend, database integration, authentication, paper management, and dashboard before integrating and evaluating the advanced AI components.

Challenges We Faced

One of the biggest challenges was combining cybersecurity with AI-based document analysis in a single system. Protecting confidential examination papers requires more than simply storing files securely; it also requires controlling access, maintaining reliable records, verifying integrity, and investigating suspicious events.

Another challenge was designing a useful leak-detection pipeline. Scanned documents may contain poor-quality images, different layouts, formatting changes, or partially altered questions. OCR and semantic similarity must work together to identify relevant content without producing too many false alarms.

Integrating multiple technologies also introduced challenges involving backend configuration, database connectivity, authentication, encryption, and communication between Java and Python services. We are addressing these issues through modular development, incremental testing, and clear separation of responsibilities.

A further challenge is evaluating detection performance realistically. A similarity score alone cannot prove that a leak has occurred. The system therefore needs carefully designed test cases, appropriate thresholds, and human verification before an alert is treated as a confirmed incident.

What We Learned

Working on ExamGuard-X has helped us understand how full-stack development, cybersecurity, and artificial intelligence can work together to solve a real-world problem.

We gained practical experience with REST API development, backend architecture, database integration, authentication, encryption, and frontend-backend communication. We also explored how OCR, semantic embeddings, vector search, and anomaly detection can support document analysis and security investigations.

Most importantly, we learned that a useful AI security system must balance detection capability, privacy, reliability, and usability. It is not enough to build a model that produces predictions; the complete system must also protect sensitive information, explain potential risks, and support responsible human decision-making.

What's Next

Our next steps are to complete the core application, integrate the AI detection pipeline, improve security event monitoring, and evaluate the system using controlled or synthetic examination documents.

We plan to measure OCR accuracy, precision, recall, F1-score, false-positive rate, detection time, and API response time. We also aim to strengthen auditability and deployment readiness through containerization and cloud infrastructure.

ExamGuard-X is our effort to make examinations more trustworthy by combining secure software engineering with AI-assisted detection and investigation.

Built With

• java