PROJECT NETRA v5.0

Neural Evaluation & Tracking Research Architecture

Multi-Modal Deepfake Detection Platform for Indian Media

Live Web App: https://netra-deepfake-detector.vercel.app/
GitHub Repository: https://github.com/sparsh101sparsh/netra-v5
Presentation: NETRA_v5_Presentation.pptx

---

🎯 Overview

PROJECT NETRA v5.0 is an AWS-cloud-first, multi-modal deepfake detection platform engineered specifically for Indian media content. It combines spatial face-swap detection, regional acoustic voice clone classification, cross-modal semantic verification, and automated AI forensic reasoning into a unified defense system.

🚀 Key Features

• EfficientNet-B4 Spatial Detector: PyTorch vision model analyzing boundary blending artifacts, facial geometry, and skin texture anomalies.
• Wav2Vec2-XLSR Audio Classifier: Mel-spectrogram voice clone detector trained on regional Indian speech (Hindi, Tamil, Telugu, Bengali).
• CLIP ViT-L/14 Context Check: Semantic alignment model verifying visual-text consistency and frame edits.
• Amazon Bedrock AI Auditor: Claude 3.5 Sonnet synthesizes multi-model ensemble weights into court-admissible PDF audit reports.
• WhatsApp Media Resilience: Robust against low-resolution compressed video artifacts.
• AWS Cloud-First Architecture: Serverless S3 + Lambda triggers + EC2 g4dn.xlarge GPU Spot compute costing ~$0.02 per video analysis.

---

👥 Team

• Sparsh (ML Lead & Architecture)
• Sudiksha (Frontend & UI Specialist)
• Sumit (Backend & Data Engineer)
• Shashwat (Forensics & Bedrock Specialist)
• Ranjan (Infrastructure & Testing)

---

🛠️ Local Setup

1. Clone the repository:

   git clone https://github.com/sparsh101sparsh/netra-v5.git
   cd netra-v5

2. Install dependencies:

   npm install

3. Start the dev server:

   npm run dev

   Open http://localhost:3000 in your browser.

---

📄 License

This project is open-source under the MIT License.