🛡️ ReviewGuard - AI-Powered Review Moderation Platform

🚀 AI-Native Moderation • Real-Time Threat Intelligence • Semantic Duplicate Detection

---

🌐 Live Demo

🚀 https://review-guard-8dcn.onrender.com/

---

📖 Overview

ReviewGuard is a production-grade AI-powered moderation platform engineered to simulate how modern digital ecosystems protect communities from:

• Toxicity
• Spam
• Duplicate reviews
• Coordinated abuse
• Manipulative content
• Repeat offenders

The system combines:

• 🧠 Large Language Models (LLMs)
• 🔍 Semantic similarity embeddings
• ⚡ Real-time WebSocket synchronization
• 🛡️ Deterministic fallback moderation
• 📡 Live telemetry architecture
• 🔐 Enterprise-grade security systems

Unlike traditional moderation dashboards, ReviewGuard operates as a complete Trust Infrastructure System with role-isolated operational layers for:

• Users
• Moderators
• Administrators

---

✨ Core Highlights

🧠 AI-Native Moderation Pipeline

• LLM-based contextual toxicity auditing
• Semantic duplicate detection
• Explainable AI reasoning
• Multi-layer moderation defense
• Fail-soft moderation fallback

---

⚡ Real-Time Event Infrastructure

• Instant moderation updates
• Live dashboard synchronization
• WebSocket-driven status propagation
• Real-time threat feeds
• Moderator queue streaming

---

🔐 Enterprise Security Layer

• JWT Authentication
• Role-Based Access Control (RBAC)
• Tiered Rate Limiting
• bcrypt password hashing
• Audit logging system
• Protected route middleware

---

📊 Analytics & Governance

• Moderator analytics dashboards
• Threat distribution insights
• Audit trail system
• Staff governance controls
• Performance leaderboards

---

🏗️ Full System Architecture

flowchart TD

    A[Landing Page]

    A --> B[JWT Authentication Gate]

    B -->|No Token| C[Login]
    B -->|No Token| D[Sign Up]
    C --> E[JWT Issued]
    D --> E

    B -->|Valid Token| F[Role Router]
    E --> F

    F --> G[User Dashboard]
    F --> H[Moderator Dashboard]
    F --> I[Admin Dashboard]

    %% User Features
    G --> G1[Submit Review]
    G1 --> G2[Trust Meter]
    G2 --> G3[Reputation Ladder]
    G3 --> G4[Review History]
    G4 --> G5[Soft Delete]

    %% Moderator Features
    H --> H1[Command Queue]
    H1 --> H2[Evaluation Studio]
    H2 --> H3[Evasion Detection]
    H3 --> H4[Audit Timeline]
    H4 --> H5[Approve Reject]

    %% Admin Features
    I --> I1[Live Threat Feed]
    I1 --> I2[Staff Governance]
    I2 --> I3[Cascading Delete]
    I3 --> I4[Global Audit Logs]
    I4 --> I5[Leaderboard]

    %% API Layer
    G5 --> J[API Gateway]
    H5 --> J
    I5 --> J

    J --> J1[JWT Verify]
    J --> J2[Rate Limiter]
    J --> J3[Socket.IO]
    J --> J4[Express.js]

    %% Core Services
    J --> K1[Review Service]
    J --> K2[Moderation Service]
    J --> K3[User Service]
    J --> K4[Admin Service]

    %% Queue
    K1 --> L[BullMQ + Redis Queue]
    K2 --> L

    %% AI Pipeline
    L --> M1[Layer 1 - Hugging Face Embeddings]
    L --> M2[Layer 2 - Groq LLM]
    L --> M3[Layer 3 - Rule Based Fallback]

    %% Storage
    M1 --> N1[MongoDB Atlas]
    M2 --> N2[Redis]
    M3 --> N3[Vector Store]

    %% Audit Flow
    K2 -. Audit Logs .-> N1
    K4 -. Decisions .-> N1

Loading

🚀 Technology Stack

Layer	Technologies
Frontend	React 18, Vite, Tailwind CSS
Backend	Node.js, Express.js
Database	MongoDB Atlas + Mongoose
Authentication	JWT + bcrypt
Real-Time	Socket.io
AI Infrastructure	Groq (Llama-3), HuggingFace
Embeddings	BGE-Small-En
Queue System	BullMQ
Broker	Redis
Charts	Recharts
Validation	express-validator
Security	express-rate-limit
Deployment	Render

---

🧠 AI Intelligence Infrastructure

🔍 Layer 1 - Semantic Duplicate Detection

Purpose

Detect semantically similar reviews even when wording changes.

Technologies

• TF-IDF Vectorization
• Semantic Embeddings
• Cosine Similarity

Capabilities

✅ Duplicate review detection
✅ Semantic spam analysis
✅ Coordinated review manipulation detection
✅ Paraphrased duplicate identification

Threshold

Similarity ≥ 0.85 → Duplicate Review

---

🤖 Layer 2 - Neural Toxicity Detection

Powered By

• Groq API
• Llama-3-8B

Detection Categories

Category	Severity
Profanity	0.35
Hate Speech	0.65
Threats	0.90
Spam	0.45
Personal Attacks	0.55

---

Advanced Signals

• Excessive CAPS
• Aggressive punctuation
• Contextual hostility
• Threat probability scoring
• Manipulative language patterns

---

Explainable AI (XAI)

Each moderation decision includes:

• AI-generated reasoning
• Confidence scoring
• Toxicity evidence
• Duplicate evidence

---

🛡️ Layer 3 - Deterministic Shield

Purpose

Fallback moderation during:

• AI outages
• API failures
• Network instability
• Queue congestion

Mechanism

• Regex heuristics
• TF-IDF similarity
• Rule-based scoring

Advantages

✅ Zero external dependency
✅ Ultra-fast execution
✅ Fail-soft protection
✅ Guaranteed moderation continuity

---

⚡ Real-Time Moderation Pipeline

User Submits Review
          ↓
Parallel AI Analysis
          ↓
Duplicate Detection
          ↓
LLM Toxicity Audit
          ↓
Fallback Shield Validation
          ↓
Threat Classification
          ↓
Moderator Queue
          ↓
Socket.io Event Emission
          ↓
Instant Dashboard Update

---

👤 User Role - Content Contributor

Features

🧠 AI Pre-Check Warnings

Users receive:

• Toxicity alerts
• Duplicate warnings
• Submission recommendations
• AI-generated moderation hints

---

📈 Trust & Impact Meter

Reputation Tiers

• Probationary
• Trusted Reviewer
• Verified Contributor
• Elite Guardian

Formula

(Approved Reviews / Total Reviews) × 100

Visual System

• Animated SVG circular progress bars
• Dynamic stroke animations
• Live trust recalculations

---

📜 Symmetric Review History

Features

• Vertically locked lifecycle feed
• Real-time review state updates
• Status accent visualization
• Responsive mirrored layout system

---

🗑️ Soft Delete Integrity Layer

Users may remove reviews publicly.

However:

• Reviews remain internally archived
• Moderators can detect evasion attempts
• Prevents delete-and-retry abuse

---

🛡️ Moderator Role - System Evaluator

---

⚡ Command Queue

• Prioritized flagged review stream
• AI severity indicators
• Duplicate confidence metrics
• Real-time synchronization

---

🔬 Evaluation Studio

• Side-by-side duplicate comparison
• AI-generated reasoning panels
• Semantic similarity analysis
• Moderation evidence visualization

---

🕵️ Evasion Detection

Moderators can inspect:

• Deleted reviews
• Repeat abuse patterns
• Bot-like review behavior
• Coordinated spam clusters

---

🧾 Audit Timeline

Every review includes:

• Submission timestamps
• AI scan completion time
• Moderator action logs
• Status transitions

---

👑 Admin Role - Strategic Overseer

---

📡 Live AI Threat Feed

Terminal-style telemetry stream displaying:

• Incoming reviews
• Threat classifications
• Moderator actions
• Queue activity
• System alerts

---

🏛️ Staff Governance Hub

Admin Capabilities

✅ Promote/Demote users
✅ Toggle account access
✅ Deactivate moderators
✅ Cascading account deletion
✅ Governance-level control

---

📚 Global Audit Logs

Complete moderation traceability including:

• Approval/Rejection logs
• Moderator identity
• Precision timestamps
• Accountability tracking

---

🏆 Staff Performance Leaderboard

Tracks:

• Review throughput
• Moderation efficiency
• Approval accuracy
• Response latency

---

⚙️ Asynchronous AI Infrastructure

BullMQ + Redis Queue System

Workflow

Producer (Submission)
         ↓
Redis Queue
         ↓
BullMQ Worker
         ↓
AI Processing
         ↓
Database Persistence
         ↓
Socket.io Push

---

Benefits

✅ Non-blocking APIs
✅ Horizontal scalability
✅ Retry resilience
✅ Queue persistence
✅ Low-latency user experience

---

🔐 Security Infrastructure

Security Layer	Implementation
JWT Authentication	7-Day Expiry
Password Hashing	bcrypt
RBAC	Route-level middleware
Validation	express-validator
Rate Limiting	Multi-tier protection
CORS Protection	Restricted origins
Audit Logging	Immutable tracking

---

🚨 Tiered Rate Limiting

Endpoint	Limit
General API	10k / 15m
Review Submission	1k / hour
Authentication	500 / 15m

---

📁 ReviewGuard Project Structure

Review Guard/
├── backend/
│   ├── config/
│   │   └── db.js              # MongoDB Atlas connection
│   │
│   ├── controllers/           # Business logic
│   │   ├── authController.js
│   │   ├── reviewController.js
│   │   ├── moderationController.js
│   │   └── adminController.js
│   │
│   ├── middleware/
│   │   ├── auth.js            # JWT verification
│   │   ├── roleCheck.js       # Role-based access
│   │   └── rateLimiter.js     # Rate limiting
│   │
│   ├── models/
│   │   ├── User.js
│   │   ├── Review.js
│   │   └── AuditLog.js
│   │
│   ├── routes/
│   │   ├── auth.js
│   │   ├── reviews.js
│   │   ├── moderation.js
│   │   └── admin.js
│   │
│   ├── scripts/
│   │   └── seed.js            # Database seeder
│   │
│   ├── utils/
│   │   ├── toxicity.js        # Mock AI toxicity detector
│   │   └── tfidf.js           # TF-IDF duplicate detection
│   │
│   ├── .env                   # Environment variables
│   └── server.js
│
└── frontend/
    └── src/
        ├── api/
        │   └── axios.js
        │
        ├── components/
        │   ├── Navbar.jsx
        │   ├── StatusBadge.jsx
        │   ├── ProtectedRoute.jsx
        │   └── LoadingSpinner.jsx
        │
        ├── context/
        │   └── AuthContext.jsx
        │
        └── pages/
            ├── Login.jsx
            ├── Register.jsx
            ├── UserDashboard.jsx
            ├── ModeratorPanel.jsx
            ├── Analytics.jsx
            └── AdminPanel.jsx

---

📡 API Endpoints

Authentication

POST   /api/auth/register
POST   /api/auth/login
GET    /api/auth/me

---

Reviews

POST   /api/reviews
GET    /api/reviews/my
DELETE /api/reviews/:id

---

Moderation

GET    /api/moderation/reviews
PUT    /api/moderation/reviews/:id/approve
PUT    /api/moderation/reviews/:id/reject
GET    /api/moderation/analytics
GET    /api/moderation/audit-logs

---

Admin

GET    /api/admin/users
PUT    /api/admin/users/:id/role
PUT    /api/admin/users/:id/toggle-status
DELETE /api/admin/users/:id

---

🧪 Test Credentials

Role	Email	Password
👑 Admin	admin@reviewmod.com	Admin@123
🛡️ Moderator	moderator@reviewmod.com	Mod@1234
👤 User	user@reviewmod.com	User@1234
👤 User	alice@reviewmod.com	Alice@1234

---

⚙️ Local Setup

Clone Repository

git clone <your-repository-url>
cd ReviewGuard

---

Backend Setup

cd backend
npm install

Create .env

MONGO_URI=your_mongodb_uri
JWT_SECRET=your_secret_key
PORT=5000
CLIENT_URL=http://localhost:5173

---

Frontend Setup

cd frontend
npm install

---

Seed Database

cd backend
npm run seed

---

Start Development Servers

Backend

npm run dev

Frontend

npm run dev

---

🎯 Key Engineering Concepts Demonstrated

• Full-Stack MERN Development
• AI-Powered Moderation Systems
• Semantic Similarity Analysis
• WebSocket Event Architecture
• Real-Time Synchronization
• Distributed Queue Systems
• Explainable AI
• Role-Based Authorization
• Scalable Backend Infrastructure
• Threat Intelligence Design

---

👩‍💻 Author

Keerthishree Kesavan 🌷

AI/ML • Full-Stack Development • Explainable AI • Real-Time Systems

---

⭐ Support

If you like this project, give it a ⭐ on GitHub and feel free to fork or contribute.