jihe520/MathModelAgent

What it solves

MathModelAgent automates the entire process of mathematical modeling for competitions. It reduces the time required to produce a high-quality modeling paper from days to approximately one hour by handling problem analysis, model design, coding, visualization, and final paper formatting.

How it works

The project operates as a set of "SKILLS" that can be driven by an agent harness (like Claude Code or Codex). It uses a multi-agent architecture consisting of specialized roles such as a modeler, coder, and writer. The workflow follows a structured pipeline:

1. Analysis: Analyzes the problem and designs the solving strategy.
2. Coding: Writes and executes code via a local Jupyter-based interpreter or cloud interpreters (E2B, Daytona) to generate results and charts.
3. Visualization: Creates technical route maps and flowcharts using DrawIO.
4. Writing: Uses Typst templates (supporting major competitions like MCM/ICM) to format the final PDF paper.
5. Verification: Performs a 9-step check for numerical consistency and compilation errors.

Who it’s for

It is designed for students and researchers participating in mathematical modeling competitions who want to automate the tedious parts of data processing and paper drafting, as well as developers looking to build specialized AI agents for scientific tasks.

Highlights

• End-to-End Automation: A single command can trigger the entire flow from problem analysis to a submission-ready PDF.
• Typst Integration: Includes 17 professional paper templates for various national and international competitions.
• Flexible Execution: Supports both local and cloud-based code interpreters.
• Human-in-the-Loop: Allows users to intervene at critical nodes to confirm, edit, or regenerate content.
• Academic Conversion: Includes a specific skill to rewrite mathematical modeling papers into English conference papers using IEEE LaTeX templates.