TruthScan AI — Fake News Detection Engine

Multi-model AI pipeline for real-time misinformation detection
7-stage RAG-powered analysis chain · Multimodal OCR · Live news scanning

Overview

TruthScan AI is a production-grade fake news detection platform that combines multiple AI models, retrieval-augmented generation (RAG), and multimodal image understanding to analyze text, URLs, and images for misinformation. The system runs a 7-stage analysis pipeline producing credibility scores, risk assessments, and evidence-based verdicts.

Live Production:

Frontend: truthscannai.netlify.app
Backend API: Hosted on Render

Architecture

┌─────────────────────────────────────────────────────────────────────┐
│                         FRONTEND (Netlify)                           │
│  React 19 · Vite 8 · Tailwind CSS · Framer Motion                  │
├─────────────────────────────────────────────────────────────────────┤
│                                                                     │
│  ┌──────────┐  ┌──────────┐  ┌──────────┐  ┌──────────────────┐   │
│  │ Detector │  │   Scan   │  │ LiveNews │  │ Reports/Profile  │   │
│  │  Page    │  │ Content  │  │  Feed    │  │   Dashboard      │   │
│  └────┬─────┘  └────┬─────┘  └────┬─────┘  └────────┬─────────┘   │
│       │              │              │                  │             │
│  ┌────▼──────────────▼──────────────▼──────────────────▼─────────┐  │
│  │              AI ORCHESTRATOR (7-Stage Pipeline)                │  │
│  │  Input → RAG Retrieval → 2× Classifiers → Sentiment →        │  │
│  │  LLM Reasoning → Credibility Engine → Verdict                 │  │
│  └───────────────────────────────────────────────────────────────┘  │
│       │                                                             │
│  ┌────▼─────────────────────────────────────────────────────────┐   │
│  │  HuggingFace Inference API  ·  NVIDIA NIM (via Netlify Fn)   │   │
│  └──────────────────────────────────────────────────────────────┘   │
└─────────────────────────────────────────────────────────────────────┘
                              │
                    REST API + WebSocket
                              │
┌─────────────────────────────▼───────────────────────────────────────┐
│                         BACKEND (Render)                             │
│  Node.js · Express 5 · MongoDB · Redis · BullMQ · Socket.IO        │
├─────────────────────────────────────────────────────────────────────┤
│                                                                     │
│  ┌────────────┐  ┌────────────┐  ┌──────────────────────────────┐  │
│  │    Auth    │  │  Payments  │  │   Image Analysis (Qwen-VL)   │  │
│  │  JWT/OTP  │  │  Razorpay  │  │  OCR + Context + Signals     │  │
│  └────────────┘  └────────────┘  └──────────────────────────────┘  │
│                                                                     │
│  ┌────────────┐  ┌────────────┐  ┌──────────────────────────────┐  │
│  │   Redis    │  │   BullMQ   │  │       Socket.IO (/ws)        │  │
│  │  Caching   │  │   Worker   │  │    Live scan updates         │  │
│  └────────────┘  └────────────┘  └──────────────────────────────┘  │
└─────────────────────────────────────────────────────────────────────┘

AI Pipeline — 7 Stages

Stage	Model / Engine	Purpose
1. Input	Qwen2.5-VL / Tesseract.js	OCR extraction + URL content fetching
2. RAG Retrieval	Custom knowledge base (8 domains)	Fact-check context grounding
3. Classifier Ensemble	RoBERTa + BERT-tiny (HuggingFace)	FAKE/REAL binary classification
4. Sentiment & Manipulation	Twitter-RoBERTa + 9 regex patterns	Emotional manipulation detection
5. LLM Reasoning	Zephyr-7B → Mistral-7B → Llama-3.1-70B	Evidence-based analysis with RAG context
6. Source Reliability	Pattern-based heuristics	Attribution, credibility markers
7. Credibility Engine	Weighted aggregation (35/15/30/20)	Final score + risk level + verdict

Multimodal Image Analysis (Backend)

Image Upload → Qwen2.5-VL (primary) → GLM-4.5V (fallback) → Legacy Chain
                                                                    │
                                              ┌─────────────────────┤
                                              │  DeepSeek-VL        │
                                              │  Florence-2 / Donut │
                                              │  TrOCR              │
                                              │  BLIP Captioning    │
                                              └─────────────────────┘

Extracts all visible text (multilingual OCR)
Generates contextual image description
Detects misinformation visual signals
SHA-256 hash-based Redis caching (7-day TTL)
BullMQ background processing for heavy images (>350KB)

Features

Detection & Analysis

Text Analysis — Paste any article, claim, or social media post
URL Analysis — Auto-fetches and extracts article content from URLs
Image/Screenshot Analysis — Camera capture, file upload, or remote URL
Live News Feed — Real-time articles from Event Registry API with one-click AI fact-checking
Pipeline Visualizer — Real-time 7-stage progress animation during analysis

User System

Email + OTP Verification — Brevo transactional email
Google OAuth — Firebase Authentication
Profile Management — Avatar upload (Cloudinary), password change, account deletion
Scan History — Full history with Redis-cached retrieval

Monetization

Freemium Model — 2 free scans, then paywall
Razorpay Integration — ₹150/month premium (30-day subscription)
Auto-expiry — Downgrades to free when subscription expires

Reports & Education

PDF Intelligence Reports — Downloadable analysis reports (jsPDF)
Awareness Section — Educational content on misinformation
Detailed Breakdowns — Layer-by-layer scoring transparency

Tech Stack

Frontend

Technology	Version	Purpose
React	19.2.5	UI framework
Vite	8.0.10	Build tool + dev server
Tailwind CSS	3.4.19	Utility-first styling
Framer Motion	12.38.0	Animations + transitions
React Router	7.15.0	Client-side routing
Firebase	12.13.0	Google OAuth
@huggingface/inference	4.13.15	Client-side AI model calls
Tesseract.js	7.0.0	Client-side OCR fallback
socket.io-client	4.8.3	Real-time WebSocket updates
jsPDF	4.2.1	PDF report generation

Backend

Technology	Version	Purpose
Node.js	≥18	Runtime
Express	5.2.1	HTTP framework
Mongoose	9.6.2	MongoDB ODM
Redis	5.12.1	Caching + rate limiting
BullMQ	5.76.8	Background job queue
ioredis	5.10.1	Redis client for BullMQ
Socket.IO	4.8.3	WebSocket server
Razorpay	2.9.6	Payment gateway
Cloudinary	1.41.3	Image CDN + uploads
Helmet	8.1.0	Security headers
express-rate-limit	8.5.1	API rate limiting
bcryptjs	3.0.3	Password hashing
jsonwebtoken	9.0.3	JWT authentication
multer	2.1.1	File upload handling

Infrastructure

Service	Purpose
Netlify	Frontend hosting + serverless functions
Render	Backend hosting
MongoDB Atlas	Database
Upstash Redis	Caching + queues + rate limiting
Cloudinary	Profile image storage
Brevo	Transactional email (OTP, welcome, payment confirmation)
HuggingFace	AI model inference (classifiers, sentiment, VLMs)
NVIDIA NIM	Llama-3.1-70B reasoning (via Netlify proxy)
Event Registry	Live news article feed

Project Structure

TruthScanAI-FakeNewsDetector/
├── backend/
│   ├── config/
│   │   ├── db.js                    # MongoDB connection
│   │   ├── redis.js                 # Redis client
│   │   └── queue.js                 # BullMQ queue bootstrap
│   ├── controllers/
│   │   ├── authController.js        # Signup, login, OTP, profile, Google OAuth
│   │   ├── scanController.js        # Save scan results, get history
│   │   ├── paymentController.js     # Razorpay order + verification
│   │   └── imageScanController.js   # Qwen2.5-VL image analysis endpoints
│   ├── middleware/
│   │   ├── authMiddleware.js        # JWT cookie verification
│   │   ├── usageMiddleware.js       # Free/premium quota enforcement
│   │   └── errorMiddleware.js       # Global error handler
│   ├── models/
│   │   ├── User.js                  # User schema (auth, subscription, scans)
│   │   ├── ScanHistory.js           # Scan result records
│   │   └── Subscription.js          # Payment/subscription records
│   ├── routes/
│   │   ├── authRoutes.js            # /api/auth/*
│   │   ├── scanRoutes.js            # /api/scan/*
│   │   └── paymentRoutes.js         # /api/payment/*
│   ├── services/
│   │   ├── qwenVLService.js         # Qwen2.5-VL multimodal OCR (primary)
│   │   ├── deepseekOCRService.js    # Legacy OCR/captioning fallback chain
│   │   └── emailService.js          # Brevo transactional email
│   ├── workers/
│   │   ├── imageAnalysisWorker.js   # BullMQ processor for heavy images
│   │   └── socket.js                # Socket.IO bridge for live updates
│   ├── utils/
│   │   ├── asyncHandler.js          # Express async error wrapper
│   │   ├── generateToken.js         # JWT cookie generation
│   │   └── razorpay.js              # Razorpay instance
│   ├── app.js                       # Express app setup (CORS, helmet, routes)
│   ├── server.js                    # HTTP server + Socket.IO + worker boot
│   └── package.json
│
├── frontend/                      # Frontend (React + Vite)
│   ├── netlify/functions/
│   │   └── nvidia-proxy.js          # Serverless NVIDIA API proxy
│   ├── public/
│   ├── src/
│   │   ├── components/
│   │   │   ├── Home.jsx             # Landing page
│   │   │   ├── DetectorPage.jsx     # Main detector (/detect)
│   │   │   ├── ScanContent.jsx      # Camera/upload scanner (/scan)
│   │   │   ├── LiveNews.jsx         # Real-time news feed + AI fact-check
│   │   │   ├── Profile.jsx          # User dashboard
│   │   │   ├── Pricing.jsx          # Plans + Razorpay checkout
│   │   │   ├── Reports.jsx          # PDF report generation
│   │   │   ├── Login.jsx / Signup.jsx
│   │   │   ├── Navbar.jsx / Footer.jsx
│   │   │   └── ...                  # Hero, BentoGrid, Awareness, etc.
│   │   ├── services/
│   │   │   ├── orchestrator.js      # 7-stage pipeline coordinator
│   │   │   ├── classifier.js        # 2× HF fake news classifiers
│   │   │   ├── sentiment.js         # Sentiment + manipulation detection
│   │   │   ├── reasoning.js         # LLM reasoning (Zephyr → Mistral → Llama)
│   │   │   ├── credibilityEngine.js # Weighted score aggregation
│   │   │   ├── retriever.js         # RAG fact-check knowledge base
│   │   │   ├── ocr.js              # Qwen-VL primary + Tesseract fallback
│   │   │   ├── huggingface.js       # Pipeline entry point + HTML formatter
│   │   │   ├── api.js              # Centralized backend API client
│   │   │   └── liveSocket.js        # Socket.IO client for live updates
│   │   ├── context/
│   │   │   └── AuthContext.jsx      # Global auth state + localStorage hydration
│   │   ├── utils/
│   │   │   └── cleanOCRText.js      # OCR text normalization
│   │   ├── App.jsx                  # Router + providers
│   │   └── main.jsx                 # Entry point
│   ├── netlify.toml                 # Netlify build + redirects config
│   ├── vite.config.js
│   ├── tailwind.config.js
│   └── package.json
│
└── README.md

API Reference

Authentication

Method	Endpoint	Auth	Description
POST	`/api/auth/signup`	—	Register with email (sends OTP)
POST	`/api/auth/verify-otp`	—	Verify email OTP → sets JWT cookie
POST	`/api/auth/login`	—	Login → sets JWT cookie
POST	`/api/auth/google`	—	Google OAuth login
POST	`/api/auth/logout`	—	Clear JWT cookie
GET	`/api/auth/me`	JWT	Get user profile (Redis cached)
PUT	`/api/auth/me`	JWT	Update profile (multipart for avatar)
POST	`/api/auth/request-delete-otp`	JWT	Request account deletion OTP
POST	`/api/auth/delete-account`	JWT	Confirm deletion with OTP
POST	`/api/auth/support`	JWT	Send support query email

Scan & Analysis

Method	Endpoint	Auth	Description
POST	`/api/scan/text`	JWT + Quota	Save scan result to history
GET	`/api/scan/history`	JWT	Get user's scan history (cached)
POST	`/api/scan/image`	—	Multimodal OCR (Qwen2.5-VL)
POST	`/api/scan/image/url`	—	Analyze remote image URL
GET	`/api/scan/image/:jobId`	JWT	Poll background job status

Payments

Method	Endpoint	Auth	Description
POST	`/api/payment/create-order`	JWT	Create Razorpay order (₹150)
POST	`/api/payment/verify-payment`	JWT	Verify + activate premium

WebSocket

Event	Direction	Description
`subscribe`	Client → Server	Join user room for updates
`image-analysis:update`	Server → Client	Background job completion

Getting Started

Prerequisites

Node.js ≥ 18
MongoDB Atlas account
Upstash Redis account
HuggingFace API token
Razorpay account (for payments)
Cloudinary account (for image uploads)
Brevo account (for transactional email)

Backend Setup

cd backend
cp .env.example .env
# Fill in all environment variables in .env
npm install
npm run dev

Frontend Setup

cd frontend
cp .env.example .env
# Fill in VITE_API_URL, VITE_NVIDIA_API_KEY, VITE_HF_TOKEN
npm install
npm run dev

Environment Variables

Backend (.env)

PORT=5001
NODE_ENV=development
MONGO_URI=mongodb+srv://...
JWT_SECRET=<openssl rand -hex 32>
CLIENT_URL=http://localhost:5173
REDIS_URL=rediss://default:...@....upstash.io:6379
CLOUDINARY_CLOUD_NAME=...
CLOUDINARY_API_KEY=...
CLOUDINARY_API_SECRET=...
RAZORPAY_KEY_ID=rzp_...
RAZORPAY_KEY_SECRET=...
BREVO_API_KEY=xkeysib-...
BREVO_FROM_EMAIL=your@email.com
BREVO_FROM_NAME=TruthScanAI
HF_TOKEN=hf_...
IMAGE_WORKER_CONCURRENCY=2

Frontend (.env)

VITE_API_URL=http://localhost:5001
VITE_NVIDIA_API_KEY=nvapi-...
VITE_HF_TOKEN=hf_...

Deployment

Frontend → Netlify

Build command: npm run build
Publish directory: dist
Node version: 20
Set environment variables in Netlify dashboard
NVIDIA proxy function auto-deploys from netlify/functions/

Backend → Render

Build command: npm install
Start command: npm start
Node version: ≥18
Set all backend environment variables
Health check: GET / returns 200

Security

JWT httpOnly cookies with SameSite=None; Secure for cross-origin production
Helmet security headers
Redis-backed rate limiting — 100 req/15min global, 10 req/15min for auth
bcrypt password hashing (10 rounds)
OTP verification for signup and account deletion
CORS whitelist — only allowed origins can make credentialed requests
Multer file filtering — only image MIME types accepted
Input validation on all endpoints

Performance

Redis caching — User profiles (1h), scan history (1h), image analysis (7d)
BullMQ background processing — Heavy images processed async with WebSocket delivery
Parallel model execution — Classifier + Sentiment run concurrently
Cascading model fallback — Never blocks on a single model failure
SHA-256 deduplication — Identical images served from cache instantly
Rate limiting — Prevents abuse without impacting legitimate users

License

ISC

Built with multiple AI models, zero tolerance for misinformation.

Navigation Menu

Repository files navigation

TruthScan AI — Fake News Detection Engine

Overview

Architecture

AI Pipeline — 7 Stages

Multimodal Image Analysis (Backend)

Features

Detection & Analysis

User System

Monetization

Reports & Education

Tech Stack

Frontend

Backend

Infrastructure

Project Structure

API Reference

Authentication

Scan & Analysis

Payments

WebSocket

Getting Started

Prerequisites

Backend Setup

Frontend Setup

Environment Variables

Deployment

Frontend → Netlify

Backend → Render

Security

Performance

License

About

Releases

Packages

Contributors

Languages

Name		Name	Last commit message	Last commit date
backend		backend
frontend		frontend
.gitignore		.gitignore
README.md		README.md

Latest commit

History

Folders and files

Repository files navigation

TruthScan AI — Fake News Detection Engine

Overview

Architecture

AI Pipeline — 7 Stages

Multimodal Image Analysis (Backend)

Features

Detection & Analysis

User System

Monetization

Reports & Education

Tech Stack

Frontend

Backend

Infrastructure

Project Structure

API Reference

Authentication

Scan & Analysis

Payments

WebSocket

Getting Started

Prerequisites

Backend Setup

Frontend Setup

Environment Variables

Deployment

Frontend → Netlify

Backend → Render

Security

Performance

License

About

Topics

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages