🛡️ Sentinel AI

Advanced Real-Time Deepfake Detection & Digital Content Authentication Platform

Built for the Kanad S.H.I.E.L.D. Hackathon — Problem Statement PS-69E9C82813D10

---

🔍 Overview

Sentinel AI is a cyber-intelligence platform designed for law enforcement, cyber investigation teams, and digital forensics analysts. It leverages OpenAI GPT-4o vision to perform deep forensic analysis of uploaded images and media, detecting synthetic generation artifacts, face swaps, GAN fingerprints, and digital manipulation at scale.

The platform provides a professional Security Operations Center (SOC)-style dashboard with real-time statistics, a cinematic verdict reveal system, and detailed per-analysis forensic reports with six independent forensic sub-scores.

---

✨ Features

🖥️ Operations Center Dashboard

• Live statistics: deepfakes detected, suspicious media, authentics confirmed, accuracy rate
• Threat breakdown chart with category distribution
• Real-time activity feed of recent analyses

📤 Upload & Analyze

• Drag-and-drop image/video/audio upload
• Base64 encoding with secure transmission
• Asynchronous AI analysis with live progress indicator
• Cinematic verdict reveal animation on completion

📋 Analysis History

• Full searchable table of all past analyses
• One-click delete for record management
• Status and verdict indicators per entry

🔬 Forensic Report

• Animated confidence gauge (0–100%)
• Six forensic sub-scores:
  • Face Consistency — facial feature coherence
  • Lighting Analysis — shadow and light source consistency
  • Compression Artifacts — natural vs. synthetic compression patterns
  • Frequency Analysis — frequency-domain anomaly detection
  • Temporal Consistency — frame-to-frame stability (video)
  • Noise Pattern — statistical noise distribution analysis
• Manipulation indicators list
• Metadata flags
• AI-generated forensic summary

---

🧠 How the AI Works

1. User uploads an image via the browser
2. The image is converted to a base64 data URL client-side
3. The API server receives the payload and immediately returns a 201 with status: analyzing
4. Asynchronously, the server sends the image to OpenAI GPT-4o vision with a detailed forensic prompt
5. GPT-4o returns structured JSON with verdict, confidence score, forensic sub-scores, and indicators
6. The frontend polls GET /api/analyses/:id every 2 seconds until status === "complete"
7. The verdict is revealed with a cinematic animation

Video and audio files receive a heuristic-based suspicious verdict with a recommendation for deeper codec-level analysis, since GPT-4o processes only image inputs.

---

🏗️ Architecture

┌─────────────────────────────────────────────────────────┐
│                      Browser (React)                    │
│  Dashboard │ Upload & Analyze │ History │ Forensic Report│
│            TanStack Query + Orval-generated hooks        │
└────────────────────────┬────────────────────────────────┘
                         │ HTTP (via proxy)
┌────────────────────────▼────────────────────────────────┐
│              Express 5 API Server (Node.js)             │
│  POST /api/analyses  →  fire-and-forget OpenAI call     │
│  GET  /api/analyses/:id  ←  polling                     │
│  GET  /api/stats/dashboard                              │
│  GET  /api/stats/recent-activity                        │
│  GET  /api/stats/threat-breakdown                       │
└────────┬───────────────────────────┬────────────────────┘
         │                           │
┌────────▼────────┐        ┌─────────▼─────────┐
│   PostgreSQL DB │        │  OpenAI GPT-4o    │
│  (Drizzle ORM)  │        │  Vision API       │
└─────────────────┘        └───────────────────┘

---

🛠️ Tech Stack

Layer	Technology
Frontend	React 19, Vite, TanStack Query, wouter, Tailwind CSS, framer-motion, Recharts
Backend	Express 5, Node.js 24, TypeScript 5.9
AI	OpenAI GPT-4o Vision (image analysis with JSON structured output)
Database	PostgreSQL 16 + Drizzle ORM
Validation	Zod v4 + drizzle-zod
API Contract	OpenAPI 3.1 + Orval codegen (React Query hooks + Zod schemas)
Package Manager	pnpm workspaces (monorepo)
Build	esbuild (server), Vite (frontend)

---

📁 Project Structure

sentinel-ai/
├── artifacts/
│   ├── api-server/          # Express API server
│   │   └── src/
│   │       ├── routes/
│   │       │   ├── analyses.ts   # Deepfake analysis routes + OpenAI
│   │       │   ├── stats.ts      # Dashboard statistics routes
│   │       │   └── health.ts     # Health check
│   │       └── app.ts
│   └── sentinel-ai/         # React frontend (Vite)
│       └── src/
│           ├── pages/
│           │   ├── Dashboard.tsx
│           │   ├── Analyze.tsx
│           │   ├── History.tsx
│           │   └── Report.tsx
│           └── App.tsx
├── lib/
│   ├── api-spec/
│   │   └── openapi.yaml     # API contract (source of truth)
│   ├── api-client-react/    # Generated React Query hooks
│   ├── api-zod/             # Generated Zod validation schemas
│   └── db/                  # Drizzle ORM schema + migrations
│       └── src/schema/
│           └── analyses.ts
└── pnpm-workspace.yaml

---

🚀 Getting Started

Prerequisites

• Node.js 20+
• pnpm 9+
• PostgreSQL 14+
• OpenAI API key with GPT-4o access

1. Clone the Repository

git clone https://github.com/YOUR_USERNAME/sentinel-ai.git
cd sentinel-ai

2. Install Dependencies

pnpm install

3. Configure Environment Variables

cp .env.example .env

Edit .env and fill in your values:

DATABASE_URL=postgresql://user:password@localhost:5432/sentinel_ai
OPENAI_API_KEY=sk-...your-key-here...
SESSION_SECRET=your-random-secret-string

4. Set Up the Database

pnpm --filter @workspace/db run push

5. Start Development Servers

# Terminal 1 — API Server (port 8080)
pnpm --filter @workspace/api-server run dev

# Terminal 2 — Frontend (port 23148)
pnpm --filter @workspace/sentinel-ai run dev

Open http://localhost:23148 in your browser.

---

🔧 Development Commands

# Full typecheck across all packages
pnpm run typecheck

# Regenerate API hooks and Zod schemas from OpenAPI spec
pnpm --filter @workspace/api-spec run codegen

# Push DB schema changes (dev only)
pnpm --filter @workspace/db run push

# Build all packages
pnpm run build

---

🌐 API Reference

Method	Endpoint	Description
POST	/api/analyses	Submit image/video for deepfake analysis
GET	/api/analyses	List all analyses
GET	/api/analyses/:id	Get single analysis (poll for completion)
DELETE	/api/analyses/:id	Delete an analysis
POST	/api/analyses/:id/reanalyze	Re-run analysis
GET	/api/stats/dashboard	Dashboard statistics
GET	/api/stats/recent-activity	Recent activity feed
GET	/api/stats/threat-breakdown	Threat category breakdown
GET	/api/healthz	Health check

Analysis Object

{
  "id": 1,
  "fileName": "suspect_image.jpg",
  "fileType": "image",
  "verdict": "deepfake",
  "confidenceScore": 94.5,
  "status": "complete",
  "forensicDetails": {
    "faceConsistency": 87.2,
    "lightingAnalysis": 91.0,
    "compressionArtifacts": 23.1,
    "frequencyAnalysis": 88.5,
    "temporalConsistency": 75.0,
    "noisePattern": 82.3
  },
  "manipulationIndicators": [
    "GAN-generated facial features detected",
    "Unnatural boundary artifacts around facial region",
    "Frequency domain shows non-natural patterns"
  ],
  "metadataFlags": ["No EXIF data present", "Unusual compression signature"],
  "summary": "High-confidence deepfake detected. The facial region exhibits classic GAN generation artifacts...",
  "createdAt": "2026-05-24T18:00:00.000Z"
}

---

🔐 Environment Variables

Variable	Required	Description
DATABASE_URL	✅	PostgreSQL connection string
OPENAI_API_KEY	✅	OpenAI API key (needs GPT-4o access)
SESSION_SECRET	✅	Random secret for session signing
PORT	Auto	API server port (default: 8080)
NODE_ENV	Auto	development or production

---

🏆 Hackathon Context

Event: Kanad S.H.I.E.L.D. — Security Hackathon for Intelligence, Encryption, Law Enforcement and Defence

Problem Statement: PS-69E9C82813D10 — Sentinel AI: Advanced Real-Time Deepfake Detection and Digital Content Authentication

Category: Startups, Industry Partners, Companies with Tech Solutions, and Innovators/Researchers

Target Users: Law enforcement agencies, cyber crime investigation units, digital forensics teams, intelligence agencies

---

📄 License

MIT License — see LICENSE for details.

---

Built with precision for national security. Every pixel analyzed. Every threat identified.