Multi-Agent Educational Transformation System

A sequential 4-agent pipeline that transforms educational documents into three distinct learning representations, demonstrating true agentic AI through cognitive task decomposition.

🎯 Overview

This system uses four specialized AI agents powered by Google Gemini to transform any educational document (PDF or text) into three learning formats:

1. Simplified Explanation - Beginner-friendly with analogies and examples
2. Visual Format - Scannable bullet points and hierarchical structure
3. Deep-Dive Analysis - Advanced technical insights and implications

🤖 The Four Agents

Agent 1: Document Analyzer

• Responsibility: Extract and structure core knowledge from input document
• Input: Raw document text
• Output: Structured knowledge graph with concepts, relationships, and complexity levels

Agent 2: Simplification Specialist

• Responsibility: Transform complex concepts into accessible explanations
• Input: Knowledge graph
• Output: Beginner-friendly explanation with analogies

Agent 3: Visual Synthesizer

• Responsibility: Restructure information into scannable formats
• Input: Knowledge graph
• Output: Hierarchical bullet points and quick-reference sections

Agent 4: Deep-Dive Expander

• Responsibility: Generate advanced analysis and extended insights
• Input: Knowledge graph
• Output: Technical depth with implications and edge cases

🚀 Quick Start

Prerequisites

• Node.js (v16 or higher)
• Gemini API key (Get one here)

Installation

1. Install dependencies:

npm install

2. Configure environment:

# Copy the example env file
copy .env.example .env

# Edit .env and add your Gemini API key
# GEMINI_API_KEY=your_actual_api_key_here

3. Start the server:

npm run dev

4. Open in browser:

http://localhost:3000

📁 Project Structure

SS/
├── agents/                      # Four AI agents
│   ├── documentAnalyzer.js     # Agent 1: Knowledge extraction
│   ├── simplificationSpecialist.js  # Agent 2: Beginner explanations
│   ├── visualSynthesizer.js    # Agent 3: Visual formatting
│   └── deepDiveExpander.js     # Agent 4: Advanced analysis
├── utils/                       # Utilities
│   ├── aiClient.js             # Gemini API wrapper
│   ├── documentParser.js       # PDF/text parsing
│   └── knowledgeGraph.js       # Knowledge graph data structure
├── routes/                      # API endpoints
│   └── api.js                  # Document processing routes
├── public/                      # Frontend
│   ├── index.html              # Main UI
│   ├── styles.css              # Premium dark mode design
│   └── app.js                  # Frontend logic
├── pipeline.js                  # Agent orchestration
├── server.js                    # Express server
├── config.js                    # Configuration
└── package.json                 # Dependencies

💡 How It Works

1. User uploads a PDF or text document
2. Agent 1 analyzes the document and builds a knowledge graph
3. Agents 2, 3, 4 run in parallel, each transforming the knowledge graph into their respective formats
4. Results are displayed side-by-side for comparison

🎨 Features

• ✅ True Agentic AI - Each agent makes autonomous decisions about structure, analogies, and depth
• ✅ Sequential Pipeline - Clear execution flow from analysis to transformation
• ✅ Premium UI - Modern dark mode design with smooth animations
• ✅ Real-time Status - Visual feedback showing agent pipeline progress
• ✅ Responsive Design - Works on desktop and mobile
• ✅ No Database Required - Stateless processing for MVP simplicity

🧪 Testing

Upload a sample educational document about any topic (e.g., "Photosynthesis", "Machine Learning", "Quantum Physics") and observe how the agents transform it into three distinct learning formats.

🏗️ Technology Stack

• Backend: Node.js + Express
• AI: Google Gemini 1.5 Flash
• Document Processing: PDF.js
• Frontend: Vanilla HTML/CSS/JavaScript
• Design: Modern dark mode with glassmorphism

📝 API Endpoints

• POST /api/process - Upload and process document
• GET /api/health - Health check

🎓 Why This is Agentic AI

1. Task Decomposition - Four distinct cognitive responsibilities
2. Autonomous Decision-Making - Each agent makes independent choices
3. Shared Context - Knowledge graph enables coordinated operation
4. Goal-Oriented - Each agent optimizes for its specific learning outcome
5. No Hard-Coded Rules - Agents adapt to document content dynamically

🚧 Future Enhancements

• Support for more document formats (DOCX, HTML)
• Interactive knowledge graph visualization
• Export to PDF/Markdown
• Multi-language support
• Custom agent configuration

📄 License

MIT

---

Built for hackathon demonstration of true agentic AI principles