<<<<<<< HEAD

KAVACH-AI

=======

Multimodal Deepfake Detection System Using Advanced Machine Learning Techniques

7df14d1 (UI enhanced)

DeepShield AI for upload-first deepfake detection
A responsive web application for analysing image, video, and audio authenticity with a FastAPI backend and a modern React frontend.

<<<<<<< HEAD ======= >>>>>>> 7df14d1 (UI enhanced)

---

Project Understanding

<<<<<<< HEAD KAVACH-AI is a web-first deepfake detection platform built around one reliable user journey: upload, analyse, review, repeat.

Multimodal Deepfake Detection System Using Advanced Machine Learning Techniques is a web-first deepfake detection platform built around one reliable user journey: upload, analyse, review, repeat.

7df14d1 (UI enhanced)

The repository was cleaned and reorganized so the active application is easy to understand:

• backend/ contains the FastAPI service, validation, model loading, and image/audio/video pipelines.
• frontend/ contains the responsive React application for upload, progress tracking, and result review.
• legacy/ contains archived experiments and older realtime/dashboard code that are no longer part of the running product.

What the active application does

• Accepts image, video, and audio uploads.
• Validates file type and file size before analysis.
• Runs a weighted deepfake scoring pipeline.
• Returns a readable result with verdict, confidence, model breakdown, warnings, waveform data, and sampled video frames when available.
• Works as a fully web application. The Chrome extension path has been removed.

---

Architecture Diagram

flowchart TB
    subgraph Clients["Client Tier"]
        Web["React Web App\nVite + Framer Motion + responsive UI"]
        Mobile["Mobile / Tablet Browser\nSame web experience"]
    end

    subgraph Backend["FastAPI Service"]
        Main["main.py\nCORS, lifespan, error handling"]
        Health["routers/health.py"]
        Analyse["routers/analyse.py"]
        Schemas["schemas/request.py + response.py"]
    end

    subgraph Detection["Detection Core"]
        Loader["models/loader.py\nstartup model registry"]
        Ensemble["models/ensemble.py\nweighted voting"]
        ImagePipe["pipelines/image_pipeline.py"]
        VideoPipe["pipelines/video_pipeline.py"]
        AudioPipe["pipelines/audio_pipeline.py"]
    end

    subgraph Support["Support Modules"]
        Config["config.py"]
        Files["utils/file_utils.py"]
        Logger["utils/logger.py"]
    end

    Web --> Main
    Mobile --> Main
    Main --> Health
    Main --> Analyse
    Analyse --> Schemas
    Analyse --> ImagePipe
    Analyse --> VideoPipe
    Analyse --> AudioPipe
    ImagePipe --> Loader
    VideoPipe --> Loader
    AudioPipe --> Loader
    Loader --> Ensemble
    Main --> Config
    Analyse --> Files
    Main --> Logger

Loading

---

Workflow Diagram

flowchart LR
    A["User opens web app"] --> B["Upload media\nimage / video / audio"]
    B --> C["Client validation\nsize + format + preview"]
    C --> D["POST /analyse"]
    D --> E["Server validation\nmime + size + temp file handling"]
    E --> F{"Media type"}
    F -->|Image| G["Image pipeline"]
    F -->|Video| H["Video pipeline\nframe sampling + optional audio extraction"]
    F -->|Audio| I["Audio pipeline"]
    G --> J["Weighted scoring"]
    H --> J
    I --> J
    J --> K["Verdict + confidence + warnings"]
    K --> L["Results page\nmodel scores, waveform, frame grid"]
    L --> M["Analyse another"]

Loading

---

Active Repository Layout

.
+-- backend/
¦   +-- main.py
¦   +-- config.py
¦   +-- routers/
¦   +-- models/
¦   +-- pipelines/
¦   +-- schemas/
¦   +-- utils/
¦   +-- requirements.txt
¦   +-- Dockerfile
+-- frontend/
¦   +-- src/
¦   ¦   +-- api/
¦   ¦   +-- components/
¦   ¦   +-- hooks/
¦   ¦   +-- pages/
¦   ¦   +-- styles/
¦   +-- package.json
¦   +-- Dockerfile
+-- docs/
¦   +-- API.md
¦   +-- INSTALL.md
¦   +-- CODEBASE_DIAGRAM.md
¦   +-- assets/banner.png
+-- legacy/
¦   +-- backend/
¦   +-- frontend/
¦   +-- root-docs/
¦   +-- scripts/
+-- docker-compose.yml

---

Tech Stack

Backend

• FastAPI
• Pydantic Settings
• Transformers
• Torch / Torchvision / timm
• OpenCV
• librosa / soundfile / scipy

Frontend

• React
• Vite
• Framer Motion
• Tailwind CSS
• Axios
• Lucide Icons

Runtime Strategy

• Primary startup path: Docker Compose
• Fallback path: Python venv + npm
• Environment bootstrapping via .env.example files

---

Recommended Setup

Option A: Docker Compose

This is the fastest and most reliable way to run the project on a fresh system.

1. Clone the repository

<<<<<<< HEAD
git clone https://github.com/abisheik687/kavach-ai.git
cd kavach-ai
=======
git clone https://github.com/abisheik687/Multimodal Deepfake Detection System Using Advanced Machine Learning Techniques.git
cd Multimodal Deepfake Detection System Using Advanced Machine Learning Techniques
>>>>>>> 7df14d1 (UI enhanced)

2. Create environment files

cp .env.example .env
cp backend/.env.example backend/.env
cp frontend/.env.example frontend/.env

Windows PowerShell alternative:

Copy-Item .env.example .env
Copy-Item backend\.env.example backend\.env
Copy-Item frontend\.env.example frontend\.env

3. Start the full stack

docker compose up --build

4. Open the application

• Web App: http://localhost:4173
• Backend API: http://localhost:8000
• API Docs: http://localhost:8000/docs

---

Manual Local Development

Use this if you do not want Docker.

1. Install prerequisites

• Python 3.11+
• Node.js 20+
• npm
• ffmpeg

2. Clone the repository

<<<<<<< HEAD
git clone https://github.com/abisheik687/kavach-ai.git
cd kavach-ai
=======
git clone https://github.com/abisheik687/Multimodal Deepfake Detection System Using Advanced Machine Learning Techniques.git
cd Multimodal Deepfake Detection System Using Advanced Machine Learning Techniques
>>>>>>> 7df14d1 (UI enhanced)

3. Create environment files

cp .env.example .env
cp backend/.env.example backend/.env
cp frontend/.env.example frontend/.env

Windows PowerShell alternative:

Copy-Item .env.example .env
Copy-Item backend\.env.example backend\.env
Copy-Item frontend\.env.example frontend\.env

4. Create and activate a Python virtual environment

python -m venv .venv

Windows PowerShell:

.\.venv\Scripts\Activate.ps1

macOS / Linux:

source .venv/bin/activate

5. Install backend dependencies

pip install --upgrade pip
pip install -r backend/requirements.txt

6. Install frontend dependencies

This repository currently ships with frontend/package-lock.json, so npm ci is the correct fast and deterministic install method.

npm ci --prefix frontend

7. Start the backend

cd backend
uvicorn main:app --reload

8. Start the frontend in a second terminal

cd frontend
npm run dev

9. Open the application

• Web App: http://localhost:4173
• Backend API: http://localhost:8000
• API Docs: http://localhost:8000/docs

---

Localhost Endpoints

• Web App: http://localhost:4173
• Backend API: http://localhost:8000
• API Docs: http://localhost:8000/docs
• Health Check: http://localhost:8000/health

---

Environment Files

The repository ships with these active templates:

• Root: .env.example
• Backend: backend/.env.example
• Frontend: frontend/.env.example

These defaults are already tuned for the active upload-first application.

---

Notes for Contributors

• The Chrome extension has been removed from the active product path.
• The older realtime and experimental surfaces are preserved under legacy/ for reference only.
• If Hugging Face model downloads are unavailable, the backend still runs using deterministic fallback scorers instead of fake placeholder outputs.
• Video audio extraction requires ffmpeg; the Docker backend image installs it automatically.

---

Documentation

• Installation Guide
• API Overview
• Architecture Diagram
• Compliance Notes
• Legacy Archive Notes