Our inspiration for MathPix came from a common frustration in the academic and scientific community - the tedious process of converting handwritten or image-based mathematical expressions into solvable digital format. Students and researchers often struggle with:

• Taking photos of equations from textbooks or handwritten notes
• Manually transcribing mathematical expressions into computer-readable format
• Switching between different tools for OCR, equation solving, and visualization
• Getting step-by-step explanations that are both rigorous and understandable

We wanted to create a seamless, end-to-end solution that could take a photo of a mathematical expression and not just recognize it, but solve it with detailed explanations.

MathPix is an integrated mathematical problem-solving platform that combines:

1. Image-to-LaTeX Conversion:

   • Accepts uploaded images containing mathematical expressions
   • Uses advanced OCR (Pix2Tex) to convert images into accurate LaTeX code
   • Provides real-time preview of the recognized expression
   • Allows manual editing of the LaTeX code for corrections

2. Multi-Engine Problem Solving:

   • Claude-powered step-by-step explanations with Python calculations
   • MATLAB integration for complex numerical computations
   • Wolfram Alpha integration for symbolic mathematics
   • Automatically formats mathematical output in LaTeX

3. Interactive User Interface:

   • Clean, intuitive Streamlit interface
   • Live LaTeX previews
   • Multiple solver options for different types of problems
   • Clear presentation of solutions and explanations

4. Comprehensive Solution Output:

   • Detailed step-by-step explanations
   • Rigorous mathematical proofs when applicable
   • Numerical results with appropriate precision
   • Visualizations and graphs where relevant

We constructed MathPix using a modern tech stack and multiple specialized components:

1. Core Technologies:

   • Python for the backend logic and integration
   • Streamlit for the web interface
   • LaTeX for mathematical typesetting

2. Key Components:

   • OCR Engine: Implemented using Pix2Tex, a specialized mathematical OCR model
   • Computation Engines:
     • Anthropic's Claude API for natural language processing and step-by-step solutions
     • MATLAB integration using Transplant for numerical computations
     • Wolfram Alpha API for symbolic mathematics
   • Frontend:
     • Streamlit components for UI elements
     • MathJax for LaTeX rendering
     • Custom file handling for image uploads

3. Integration Layer:

   • Custom wrapper classes for each computation engine
   • Unified error handling and output formatting
   • Asynchronous execution for responsive UI
   • Temporary file management for secure operation

4. Development Process:

   • Started with core OCR functionality
   • Added step-by-step solution generation using Claude
   • Integrated MATLAB for numerical computations
   • Added Wolfram Alpha for symbolic mathematics
   • Implemented unified LaTeX output formatting
   • Created clean, intuitive user interface
   • Added error handling and input validation
   • Optimized performance and resource usage

The project represents a careful balance of different technologies, each chosen for its strengths:

• Claude for natural language understanding and explanation generation
• MATLAB for efficient numerical computations
• Wolfram Alpha for symbolic mathematics
• Streamlit for rapid UI development
• LaTeX for professional mathematical typesetting

pip install requirements.txt
streamlit run app.py

• Alex Feng
• David Chang

Licensed under the MIT License.