An intelligent multi-agent system for detecting AI-generated deepfake images, powered by Google Gemini and built for the Kaggle Agents Intensive Capstone Project.
The Challenge: The proliferation of AI-generated deepfake images poses serious threats to information integrity, personal privacy, and public trust. Manual detection is time-consuming and requires expertise.
The Solution: An AI agent system that not only detects deepfakes but also explains the results in natural language, answers follow-up questions, and maintains conversation context for iterative analysis.
Traditional deepfake detection tools provide binary outputs without explanation. This multi-agent system offers:
- 🤖 Automated Detection - ML model analyzes images automatically
- 💬 Intelligent Explanation - LLM-powered agent explains results in plain language
- ❓ Interactive Q&A - Users can ask follow-up questions about detections
- 🧠 Context Awareness - Conversation memory enables multi-turn interactions
- 📊 Observability - Full logging and tracing for transparency
graph TD
A[User] -->|Image + Query| B[Orchestrator Agent]
B -->|Image Path| C[Detection Agent]
C -->|Run ML Model| D[DeepfakeDetectionTool]
D -->|Results| C
C -->|Detection Results| B
B -->|Results + Context| E[Analysis Agent - Gemini LLM]
E -->|Explanation| B
B -->|Response| A
B <-->|Store/Retrieve| F[Conversation Memory]
style B fill:#4CAF50
style C fill:#2196F3
style E fill:#FF9800
style F fill:#9C27B0
Agent/
├── 🐍 agent_system.py # Multi-agent orchestrator (CLI)
├── 🌐 app.py # Flask web application
├── 🔍 deepfake_detector.py # ML model wrapper
├── ⚙️ config.py # Configuration management
├── 📋 requirements.txt # Python dependencies
├── 🔒 .env.example # Environment template
├── 📖 README.md # This file
├── 📁 static/ # Web app assets
│ ├── 🎨 css/style.css # Styling
│ ├── ⚡ js/script.js # Frontend logic
│ └── 🖼️ images/ # UI assets
├── 📁 templates/ # Flask templates
│ └── 🏠 index.html # Main interface