Automate math modeling competition papers from problem to submission
You can complete the process of analyzing problems, writing code, and drafting papers during a competition with a single command. This tool takes a given problem, builds a model, runs code to extract data, and automatically completes a paper in a submittable format. The user's only tasks are to enter a model API key and run the start command.
Automation from problem analysis to paper submission
The tool begins with problem analysis, then constructs a mathematical model, writes and executes necessary code, and fixes errors on its own. Each stage is separated into an independent function, allowing you to run the entire flow or selectively use specific stages. The final output is a paper formatted in Typst, which is converted to PDF and ready for immediate submission. It includes 17 official templates for major competitions such as CUMCM, MCM/ICM, and others, automatically applying the format according to the competition type.
Verification and user intervention to maintain output quality
To ensure the quality of automatically generated results, it undergoes a 9-step automatic verification process. It checks for text omissions, numerical inconsistencies, compilation errors, and PDF visual confirmation to reduce basic mistakes. It also supports a 'human-machine collaboration' mode that waits for user approval at critical decision points. In this mode, users can intervene in the output through six options: confirm, modify, regenerate, ask questions, skip, or stop.
From desktop app to local deployment
The easiest method is to download the desktop version. Installers for macOS and Windows are provided, requiring no separate development environment setup such as Python, Node.js, or Redis. After installation, you can use it immediately by entering only the model API key. Developers or users needing customization can deploy via Docker or set up a local environment with Python, Node.js, and Redis to run it directly. For local deployment, you can start the service with the docker-compose up command or run the backend and frontend separately.
Realistic limitations of an experimental project
This project is currently in the experimental and demo stage, and many parts need improvement. The README states that AI-generated content is for reference only and that winning awards in major competitions like CUMCM is impossible at the current level. The license allows only free personal use and prohibits commercial use. Additionally, the Windows installer is unsigned, so security warnings may appear; it should be downloaded only from official channels. While this tool automates the entire math modeling process, the accuracy and creativity of the final output still rely heavily on user review and intervention.
