Skip to content

[invasion_off_1]: Multimodal Deepfake Detection & Trust Engine #15

Description

@Anushkas147

Team Name

NeuralNomads

Hackathon Track

AI/ML

ps-number

4

Email Address

anushka.shinde@vit.edu.in

Email Addresses of Team Members

dipankar.pimple@vit.edu.in , vedant.patole@vit.edu.in , kanchan.chavan@vit.edu.in

Project Description

The project I created is a Multimodal Deepfake Detection & Trust Engine — an AI-powered system that detects deepfakes across video, audio, and image formats and provides a unified trust score.

Our system goes beyond traditional single-modality detection by combining:

  • Video deepfake detection (facial inconsistencies)
  • Audio analysis (synthetic voice detection)
  • Audio-visual synchronization (lip-sync mismatch detection)
  • Image deepfake detection

The output is not just a prediction, but a complete trust analysis, including:

  • Trust Score (0–100)
  • Risk classification (Safe / Suspicious / Fake)
  • Frame-level timeline showing where manipulation occurs
  • Explainable AI insights
  • Visual heatmaps highlighting manipulated regions

The goal is to build a user-friendly trust verification system that helps people identify fake content quickly and confidently before believing or sharing it.

Inspiration behind the Project

The reason we chose this idea is the rapid rise of deepfake technology and its serious impact on misinformation, fraud, and digital trust.

With advancements in generative AI, fake videos and voices are becoming increasingly realistic and harder to detect. This has led to real-world issues such as:

  • Financial scams using voice cloning
  • Fake news and misinformation
  • Identity impersonation

Most existing solutions are either too technical, limited to one modality, or lack clear explanations.

We wanted to create a system that is:

  • Multimodal (more accurate)
  • Fast and accessible
  • Explainable and user-friendly

Our vision is to build a “trust layer for digital media”, where users can verify authenticity before trusting any content.

Tech Stack

The technologies used in this project include:

🧠 AI/ML

  • PyTorch (deepfake detection models like EfficientNet/MesoNet)
  • Librosa (audio feature extraction - MFCC)
  • MediaPipe (facial landmarks for lip-sync detection)
  • OpenCV (video processing and frame extraction)

⚙️ Backend

  • FastAPI (API development and ML orchestration)
  • Python (core backend logic)

🎨 Frontend

  • React.js (UI development)
  • Tailwind CSS (modern responsive styling)
  • Chart.js / Recharts (timeline visualization)

Database (optional MVP)

  • PostgreSQL (for storing analysis results as history)

🔧 Tools

  • Git & GitHub (version control)
  • Postman (API testing)

The system follows a modular architecture, where the frontend interacts with a FastAPI backend, which runs the ML pipeline and returns structured results.

Project Repo

https://github.com/Dipankar2105/Invasion

Demo Video

https://youtu.be/llyUoWZc5G4

Presentation Link

https://prezi.com/p/f72cj0luayx7/deepfake-trust-engine/

Anything Else?

This project was built as a functional MVP within a limited hackathon timeframe, focusing on delivering a strong combination of AI capability and user experience.

Key highlights of our solution:

  • Multimodal detection (video + audio + image)
  • Explainable AI outputs (not just predictions)
  • Interactive and intuitive dashboard
  • Real-time trust scoring system

Future improvements we aim to work on include:

  • Real-time detection for live video calls
  • Browser extension for instant verification
  • Advanced biological signals (heartbeat detection)
  • Integration with content authenticity standards (C2PA)

We believe this project can evolve into a real-world product that helps combat misinformation and build trust in digital media.

Thank you for the opportunity!

Rules and Code of Conduct

  • I agree to follow this hackathon's Rules and Code of Conduct

Metadata

Metadata

Assignees

No one assigned

    Type

    No type
    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions