🛡️ Sentinel: On-Chain Verification Hub

Sentinel is a decentralized, full-stack Web3 platform designed to combat misinformation and deepfakes by leveraging community consensus and immutable blockchain anchoring.

Whenever community consensus is reached on a piece of content (Post or Comment), the system natively generates a cryptographic SHA-256 hash of the content and permanently anchors the verdict to the Sepolia Ethereum network via a gasless relayer.

---

🛠 Tech Stack

• Frontend: React (Vite), Tailwind CSS, ethers.js, Lucide Icons.
• Backend: Node.js, Express.js, native crypto (SHA-256), ethers.js relayer.
• Smart Contract: Deployed on Sepolia Testnet.

---

🚀 Getting Started

Follow these instructions to run the full-stack environment locally for testing.

Prerequisites

1. Node.js (v18+ recommended)
2. MetaMask browser extension installed.

1. Environment Variables

Your teammate will provide you with .env files or you can use the .env.example templates.

• Backend: Place the provided .env file directly into the /backend directory. It contains the RPC_URL, RELAYER_PRIVATE_KEY, and CONTRACT_ADDRESS needed for on-chain transactions.
• AI Orchestrator: Add GOOGLE_API_KEY, GROQ_API_KEY, and OPENAI_API_KEY to /ai-orchestrator/.env.
• Moderation Service: Add HF_TOKEN (Hugging Face token) to /moderation_service/.env for downloading gated models like NudeNet and SwinV2.

2. Install Dependencies

Open four terminal windows.

Terminal 1 (Backend - Node.js):

cd backend
npm install

Terminal 2 (Frontend - React):

cd frontend
npm install

Terminal 3 (AI Orchestrator - Python):

cd ai-orchestrator
python -m venv venv
source venv/Scripts/activate

pip install -r requirements.txt
# Requires GOOGLE_API_KEY, GROQ_API_KEY, OPENAI_API_KEY in .env

Terminal 4 (Moderation Service - Python):

cd moderation_service
python -m venv venv
source venv/Scripts/activate

pip install -r requirements.txt
# Note: First run will download ~2GB of ML models (SwinV2, NudeNet). Ensure HF_TOKEN is in .env.

3. Run the Development Servers

Start all four servers simultaneously in their respective terminals:

Backend:

cd backend
npm run dev
# Expected output: 🚀 Sentinel Node.js Backend running on port 8000

Frontend:

cd frontend
npm run dev
# Expected output: ➜  Local: http://localhost:5173/

Moderation Service:

cd moderation_service
source venv/Scripts/activate

python app.py
# Expected output: INFO:     Uvicorn running on http://0.0.0.0:8002

AI Orchestrator:

cd ai-orchestrator
source venv/Scripts/activate

uvicorn app:app --host 0.0.0.0 --port 5000 --reload
# Expected output: Uvicorn running on http://0.0.0.0:5000

4. How to Close the Project

To safely shut down the servers, navigate to each of the four terminal windows you opened and press:

Ctrl + C (Windows/Linux) or Cmd + C (Mac)

If you are prompted with Terminate batch job (Y/N)? on Windows, simply type Y and press Enter. Once you have done this in all four terminals, the entire project will be closed.

---

🤖 Text Safety Engine (AI Orchestrator)

The platform features a 3-tier LangChain Fallback Hierarchy (Gemini 2.0 Flash → Groq Llama 3.3 70B → OpenAI GPT-4o-mini) that evaluates text posts in a single pass to provide:

1. AI Generation Detection: Identifies stylometric markers indicating LLM authorship.
2. Harm & Safety Assessment: Flags hate speech, misinformation, and other high-risk content using a strictly enforced 10-Tag Harm Taxonomy (e.g., Harassment, Self-Harm, Scams) and a Zero-Tolerance Rubric.
3. Domain Classification: Categorizes the text by topic (e.g., Politics, Tech).

The backend automatically routes text posts to this engine and renders the resulting scores and tags seamlessly in the frontend UI.

---

👁️ Visual Moderation & Deepfake Engine (Moderation Service)

The platform features a multi-stage image analysis pipeline that processes visual content in a single pass:

1. NSFW Detection: Uses NudeNet to identify and block explicit content before further processing.
2. C2PA Provenance: Scans for embedded Content Credentials (C2PA manifests) to detect cryptographically disclosed AI generation.
3. Deepfake & AI Generation Detection: Employs a SwinV2 Transformer (converted to ONNX FP16 for CPU efficiency) for whole-image classification, flagging deepfakes and AI-generated imagery without relying on face detection.

This service operates sequentially for early exits and uses ONNX Runtime to provide extremely fast inference.

---

🧪 Testing the "Proof of Integrity" Flow

Once the servers are running, open http://localhost:5173/dashboard in your browser.

1. Connect Wallet: Click the top right button to connect your MetaMask.
2. Submit Content: Go to the Verification Feed tab and submit a test post.
3. Trigger Consensus:
   • A post requires exactly 3 unique votes to hit consensus.
   • Vote on the post with your current wallet.
   • Switch accounts in MetaMask (Account 2, Account 3) and cast the remaining votes.
4. On-Chain Anchoring (The 15-Second Delay):
   • Upon the 3rd vote, the backend will compute the hash and call the Sepolia blockchain.
   • Wait roughly 15 seconds for the Ethereum block to be mined.
5. Verify Cryptography:
   • The post will move to the On-Chain Record tab.
   • Click the "Verify Local Integrity" button. The browser will hash the raw text locally using crypto.subtle and compare it to the blockchain record. You should see a massive green success notification!
   • Click View Record to see the actual anchored transaction on the Sepolia Blockscout explorer.

---

Built for the Hackathon. 🚀