ExamGuard AI

Intelligent Assessment Platform with RAG-Powered Question Generation & Multi-Agent Cheating Detection

Smarter Exams. Honest Results.

ExamGuard AI is a dual-purpose academic platform that automatically generates high-quality, discipline-specific exam questions from course materials using RAG, and monitors exam integrity through privacy-conscious multi-agent proctoring that analyzes behavior patterns — reducing false positive rates from the industry standard of 71% to under 5%.

Features · Tech Stack · Architecture · Repositories · Team · Getting Started

Online education faces a dual crisis:

• Professors spend 10–15 hours creating each exam while struggling to generate high-quality, varied questions
• 68% of students admit to cheating on online exams
• Existing proctoring tools (e.g., Proctorio) have a 71% false positive rate, causing student anxiety and unfair accusations

ExamGuard AI addresses both sides of the problem simultaneously:

1. AI Exam Generation — Upload course materials (PDFs, slides, syllabi) and let the RAG pipeline generate 50 exam questions across 4 cognitive levels in minutes
2. Multi-Agent Proctoring — Real-time behavioral monitoring with probabilistic reasoning that is bias-aware, explainable, and always human-reviewed before any action is taken

• RAG-Powered Question Generation — Upload any course material and get exam-ready questions instantly, mapped to Bloom's Taxonomy cognitive levels (Recall 20% / Application 30% / Analysis 30% / Synthesis 20%)
• AI Exam Builder — Drag-and-drop question editor with cognitive level distribution visualization, difficulty tuning, and question bank management
• Live Proctoring Monitor — Real-time dashboard showing all active students, integrity scores, and a live alert feed for flagged events
• Incident Review System — Timestamped event timeline, student rebuttal panel, and one-click escalation workflow
• Results & Analytics — Score distributions, per-student integrity breakdowns, and exportable reports
• Course & Material Management — Organize uploads by course, semester, and subject for reusable question banks

• Distraction-Free Exam Mode — Fullscreen-enforced, focused exam interface with real-time timer
• System Pre-Check — Camera, microphone, and fullscreen verification before every exam session
• Transparent Integrity Reports — If flagged, students receive a full explainable report with the exact event, timestamp, and confidence score
• Rebuttal Submission — Students can submit contextual explanations (e.g., accommodation needs) before a professor reviews any incident
• Exam History & Results — Full history of past exams, scores, and integrity statuses

• Platform-Wide Oversight — Manage all professors, students, exams, and escalated violations from a single admin console
• Bias Audit Reports — Demographic and accessibility accuracy reports to ensure equitable treatment across all student populations
• System Health Monitoring — Real-time logs, agent performance metrics, and uptime tracking
• Sensitivity Configuration — Institution-wide proctoring threshold controls (Strict / Balanced / Lenient)

┌─────────────────────────────────────────────────────────────────┐
│                        CLIENT (Browser)                         │
│                                                                 │
│   React 19 + TypeScript + Vite                                  │
│   ┌──────────────┐  ┌──────────────┐  ┌──────────────────────┐  │
│   │  Exam Builder│  │  Exam Taking │  │  Live Monitor        │  │
│   │  (Professor) │  │  (Student)   │  │  (Professor)         │  │
│   └──────────────┘  └──────┬───────┘  └──────────────────────┘  │
│                             │                                   │
│   ┌─────────────────────────▼──────────────────────────────┐    │
│   │              Proctoring Agents (In-Browser)            │    │
│   │  TensorFlow.js (Gaze) · Web Audio API · Visibility API │    │
│   └─────────────────────────┬──────────────────────────────┘    │
└─────────────────────────────┼───────────────────────────────────┘
                              │ REST + WebSocket (Socket.io)
┌─────────────────────────────▼───────────────────────────────────┐
│                        SERVER (NestJS)                          │
│                                                                 │
│   ┌──────────────┐  ┌──────────────┐  ┌──────────────────────┐  │
│   │  Auth Module │  │  Exam Module │  │  Proctoring Module   │  │
│   │  (JWT/Guards)│  │  (CRUD+AI)   │  │  (Events+Scoring)    │  │
│   └──────────────┘  └──────┬───────┘  └──────────────────────┘  │
│                             │                                   │
│   ┌─────────────────────────▼──────────────────────────────┐    │
│   │              AI Orchestration (LangChain.js)           │    │
│   │         Gemini Pro 3 · Multi-Agent Pipeline · RAG      │    │
│   └─────────────────────────┬──────────────────────────────┘    │
└─────────────────────────────┼───────────────────────────────────┘
                              │
┌─────────────────────────────▼───────────────────────────────────┐
│                          DATA LAYER                             │
│                                                                 │
│   MongoDB (Documents)  ·  ChromaDB (Vectors)  ·  Redis (Cache)  │
└─────────────────────────────────────────────────────────────────┘

This GitHub organization contains the following repositories:

The platform covers 39 screens across 4 user contexts:

• Node.js v20 LTS
• pnpm v9+
• Docker & Docker Compose
• MongoDB, Redis (or use the provided docker-compose.yml)

# Frontend
git clone https://github.com/examguard-ai/examguard-frontend.git

# Backend
git clone https://github.com/examguard-ai/examguard-backend.git

# From the root of any repo
docker-compose up -d

cd examguard-frontend
pnpm install
cp .env.example .env.local
pnpm dev

cd examguard-backend
pnpm install
cp .env.example .env
pnpm start:dev

# Frontend (.env.local)
VITE_API_URL=http://localhost:3000
VITE_WS_URL=ws://localhost:3000
VITE_APP_ENV=development

# Backend (.env)
PORT=3000
MONGODB_URI=mongodb://localhost:27017/examguard
REDIS_URL=redis://localhost:6379
CHROMA_URL=http://localhost:8000
JWT_SECRET=your-secret-here
JWT_EXPIRES_IN=7d
GEMINI_API_KEY=your-gemini-key-here

• Project planning & architecture design
• UI/UX design system (Stitch + shadcn/ui)
• Frontend scaffolding (React 19 + Vite + TypeScript)
• Auth module (NestJS + JWT + Passport)
• Exam CRUD & builder (Frontend + Backend)
• RAG pipeline (ChromaDB + LangChain.js + Gemini)
• AI question generation agents
• In-browser proctoring agents (TensorFlow.js + MediaPipe)
• Real-time WebSocket proctoring monitor (Socket.io)
• Integrity scoring & incident reporting
• Student rebuttal system
• Analytics & reporting dashboard
• Bias audit & accessibility testing (100-student validation study)
• Production deployment

ExamGuard AI is built with fairness and transparency as first-class requirements:

• No Punitive Automation — The AI never makes punitive decisions. All flags are reviewed by a human professor before any action is taken
• Personalized Baselines — Each student's behavior is calibrated individually during the first 5 minutes of an exam, not compared to a global average
• Explainable Flags — Every flagged incident includes the exact timestamp, event type, and confidence score, written in plain language
• Rebuttal System — Students always have the opportunity to submit context before a professor reviews any incident
• Bias Auditing — Continuous demographic and accessibility stress tests on all ML models to ensure equitable accuracy across diverse student populations
• Disability Accommodations — Professors can disable specific proctoring signals (e.g., gaze tracking) for students with documented accessibility needs

This project is licensed under the MIT License — see the LICENSE file for details.