TrustSphere AI Demo Video

• You will regret missing this demo video! Click on the thumbnail below to watch it now!

Presenting TrustSphere AI: An LLM-powered Trust & Safety platform for detecting review fraud, counterfeit listings, and seller manipulation in e-commerce.

---

What is TrustSphere AI?

TrustSphere AI is an explainable, scalable, and intelligent platform built to combat fake reviews, counterfeit products, and fraudulent seller networks in e-commerce ecosystems.
We leverage LLMs, Computer Vision, Graph Intelligence, and Explainable AI to deliver real-time trust insights to platforms, buyers, and compliance teams.

---

Key Features

✅ Review Analysis with LLMs + LangChain

• Analyze sentiment, detect AI-generated content, and flag suspicious reviews
• Uses similarity search (FAISS + embeddings) to detect review spamming
• Tech Stack: LangChain, HuggingFace Transformers, FAISS, Kafka, AWS Lambda, Pinecone, AWS SageMaker

💡 How it scales?

• Uses Pinecone for fast vector similarity lookups
• Batched and async LLM inference with LangChain + SageMaker endpoints
• Review ingestion is streamed via Kafka Streams to ensure real-time processing

🖼 CV-Based Product Listing Verification

• Identify counterfeit packaging via product image analysis
• Detects reused images, manipulated branding, and fake barcodes
• Siamese networks + Grad-CAM to highlight mismatched or reused assets
• Tech Stack: PyTorch, OpenCV, Mobile-Net, Grad-CAM, Scikit-learn, Flask, OCR, SageMaker, AWS Lambda, Amazon S3, CloudFront

💡 How it scales?

• Images are stored on Amazon S3 and served via CloudFront
• Inference runs on GPU-backed SageMaker endpoints with auto-scaling
• Precomputed visual embeddings reduce real-time load

🕸 Graph-Based Seller Networks

• Builds seller-buyer-review graphs to detect review farms and fraud rings
• GNN-powered fraud ring detection across millions of user-product interactions
• IP clustering and suspicious co-reviewing behavior modeling
• Tech Stack: PyTorch Geometric, Graph Neural Networks, Louvain Community Detection, Scikit-learn, Kafka, AWS SageMaker

💡 How it scales?

• Scheduled GNN training using SageMaker + GPU instances
• Real-time edges can be streamed into the graph via Kafka

📊 Trust Score Engine

• Calculates seller-level integrity scores based on:
  • Review quality and quantity
  • Return rates and dispute frequency
  • GNN anomaly signals and image-based inconsistencies
• Powers moderation decisions and visibility rankings
• Tech Stack: Python, Scikit-learn, Pandas, NumPy, Flask (under development)

💡 How it scales?

• Uses Redis for fast in-memory trust score caching
• Trust score logic is stateless and runs on AWS Lambda containers

🔍 Explainable AI Layer

• Review decisions are explained using LIME for LLM-based flags
• Image-based flags are explained using Grad-CAM visualizations

---

Final Architecture Overview

---

Project Structure

📦 TrustSphere-AI/
├── server/                 # Flask backend app
│   ├── models/             # ML models, embeddings, CV, GNN
│   ├── services/           # Data processing
│   ├── routes/             # API endpoints
│   ├── index.js            # Main server entry point
│   └── ...
├── client/                 # Frontend React app
│   ├── src/                # React components, pages, styles
│   ├── public/             # Static assets, icons, images
│   ├── package.json        # Frontend dependencies
│   └── ...
├── assets/                # Images, logos, and other static assets
├── README.md              # You're here!
└── requirements.txt
└── package.json

---

Instructions to Run the Project Locally

1. Clone the Repo

git clone https://github.com/HitG010/TrustSphere-AI.git
cd TrustSphere-AI

2. Set Up Virtual Environment

python -m venv venv
source venv/bin/activate   # or venv\Scripts\activate on Windows

• Or use Conda if preferred:

conda create -n trustsphere python=3.9
conda activate trustsphere

3. Install Requirements

pip install -r requirements.txt

4. Start FAISS Embedding Indexer

Make sure to run the embedding loader:

python indexing/index_reviews.py   # Pre-load review.json embeddings

5. Run the Frontend

cd frontend
npm install
npm run dev

6. Start the Node.js Backend Server

cd backend
npm install
nodemon index.js

7. Run the Flask API Service

python3 server/models/app.py

or

python server/models/app.py

And then naivgate to http://localhost:5173 in your browser to access the TrustSphere platform.

---

👥 Team

TrustSphere AI was built by a passionate team FigureOut during the Amazon HackOn'25 Hackathon.

• Kartik Bindra (Team Leader) LinkedIn | GitHub

• Hitesh Gupta LinkedIn | GitHub

• Piyush Bansal LinkedIn | GitHub