Math Agentic RAG - Intelligent Math Problem Solver

An advanced AI-powered mathematics tutoring system that combines knowledge base search, web search, and LLM capabilities to provide comprehensive step-by-step solutions to mathematical problems.

🎯 Overview

Math Agentic RAG is a sophisticated educational platform that leverages multiple AI technologies to solve mathematical problems with detailed explanations. The system intelligently routes queries through different knowledge sources to provide the most accurate and comprehensive solutions.

Key Features

🧠 Multi-Source Intelligence: Combines knowledge base, web search, and Gemini LLM for optimal results
📚 Step-by-Step Solutions: Detailed explanations with mathematical reasoning
⚡ Intelligent Routing: Automatically selects the best source based on confidence scores
🎨 Beautiful Math Rendering: LaTeX-powered mathematical notation with KaTeX
📊 Analytics & Feedback: Built-in performance tracking and user feedback system
🔒 Input Validation: Guardrails for safe and appropriate content
🚀 Real-time Processing: Fast response times with optimized caching

🏗️ Architecture

┌─────────────────┐    ┌─────────────────┐    ┌─────────────────┐
│   Frontend      │    │    Backend      │    │   External      │
│   (React)       │◄──►│   (FastAPI)     │◄──►│   Services      │
│                 │    │                 │    │                 │
│ • Search UI     │    │ • API Routes    │    │ • Qdrant DB    │
│ • Math Render   │    │ • Services      │    │ • Gemini LLM   │
│ • KaTeX Display │    │ • Guardrails    │    │ • MCP Service  │
└─────────────────┘    └─────────────────┘    └─────────────────┘

Technology Stack

Frontend:

React 18 with modern hooks
Tailwind CSS for styling
KaTeX for mathematical rendering
React Hot Toast for notifications

Backend:

FastAPI for high-performance APIs
Pydantic for data validation
Structlog for comprehensive logging
Python 3.11+

AI & Data:

Qdrant vector database for semantic search
Google Gemini 2.0 Flash for LLM processing
Model Context Protocol (MCP) for web search
Sentence Transformers for embeddings

🚀 Quick Start

Prerequisites

Python 3.11 or higher
Node.js 18+ and npm
Git

Environment Setup

Clone the repository:

git clone https://github.com/bhoomika-254/Math-Agentic-RAG.git
cd Math-Agentic-RAG

Set up environment variables:

# Create .env file in the root directory
touch .env

Add the following variables to your .env file:

# Required API Keys
GEMINI_API_KEY=your_gemini_api_key_here
QDRANT_URL=your_qdrant_cluster_url
QDRANT_API_KEY=your_qdrant_api_key
QDRANT_COLLECTION=collection_name

Backend Setup

Navigate to backend directory:

cd backend

Create and activate virtual environment:

python -m venv venv

# On Windows:
venv\Scripts\activate
# On macOS/Linux:
source venv/bin/activate

Install dependencies:

pip install -r requirements.txt

Start the backend server:

python main.py

The backend will be available at http://localhost:8000

Frontend Setup

Navigate to frontend directory (in a new terminal):

cd frontend

Install dependencies:

npm install

Start the development server:

npm start

The frontend will be available at http://localhost:3000

Database Setup

Navigate to database directory:

cd database

Install dependencies:

pip install -r requirements.txt

Ingest sample math data:

python ingest.py

📖 Usage Guide

Basic Usage

Open the application in your browser at http://localhost:3000
Enter a math question in the search box, such as:
- "Find the derivative of f(x) = 3x² + 2x - 1"
- "Solve the quadratic equation: x² + 5x + 6 = 0"
- "What is the integral of sin(x) dx?"
View the solution with step-by-step explanations and beautiful mathematical notation
Provide feedback using the thumbs up/down buttons to help improve the system

Advanced Features

Copy solutions to clipboard using the copy button
Share results with the share functionality
View metadata including confidence scores and processing time
Explore related solutions when multiple results are available

🛠️ API Documentation

The backend provides RESTful APIs accessible at http://localhost:8000/docs (Swagger UI).

Main Endpoints

POST /api/search - Submit math questions for solving
POST /api/feedback - Submit user feedback on solutions
GET /api/health - Health check endpoint

System Configuration

The system uses a three-tier approach for solving math problems:

Knowledge Base Search (Primary): High-confidence semantic search in Qdrant
Web Search via MCP (Secondary): Real-time web search for current information
Gemini LLM (Fallback): Direct AI problem solving when other sources have low confidence

Confidence Thresholds

High Confidence: ≥ 0.8 (Use KB results)
Medium Confidence: 0.6-0.79 (Use MCP results)
Low Confidence: < 0.6 (Use Gemini LLM)

🧪 Development

Project Structure

Math-Agentic-RAG/
├── backend/               # FastAPI backend
│   ├── main.py           # Application entry point
│   ├── routes/           # API route handlers
│   ├── services/         # Business logic services
│   ├── models/           # Pydantic models
│   └── requirements.txt  # Python dependencies
├── frontend/             # React frontend
│   ├── src/
│   │   ├── components/   # Reusable components
│   │   ├── pages/        # Page components
│   │   └── utils/        # Utility functions
│   ├── public/           # Static assets
│   └── package.json      # Node dependencies
├── database/             # Database setup and ingestion
│   ├── ingest.py         # Data ingestion script
│   ├── qdrant_manager.py # Database management
│   └── utils.py          # Database utilities
└── README.md             # This file

🔒 Security & Guardrails

Input validation prevents malicious content
Output sanitization ensures safe responses
Rate limiting prevents abuse
API key protection secures external services

🤝 Acknowledgments

Qdrant for vector database capabilities
Google Gemini for advanced language modeling
KaTeX for beautiful mathematical rendering
FastAPI for high-performance backend framework
React for modern frontend development

Built with ❤️ by bhoomi