WSC 2026 Tutorial Practice: Seamless Integration of Generative AI and Simulation Modeling

This article introduces the deep integration scheme of generative AI and simulation modeling proposed by Dehghanimohammadi et al. at WSC2026. The supporting open-source browser-side application is implemented based on Pyodide, allowing users to experience the AI-driven full simulation process (problem definition → input modeling → model creation → execution → experimental analysis) without installing any software. It also demonstrates the application of Agentic AI (MCP protocol), lowering the threshold for simulation modeling and pointing out the future direction of the field.

Traditional simulation modeling processes require professional knowledge, tedious programming, and manual debugging, which have a high threshold. The core objective of this project is to demonstrate how generative AI runs through the entire lifecycle of simulation modeling (problem definition, input modeling, model creation, execution, experimental analysis). The team built a browser application based on Pyodide, which compiles the Python runtime into WebAssembly, enabling users to run Python and scientific computing libraries (such as NumPy and SciPy) in the browser without installing software.

The application adopts a pure front-end architecture (HTML/CSS/native JS) and is hosted on GitHub Pages to ensure portability and zero deployment cost. Core components include:

1. Pyodide runtime: Loads libraries like scipy and numpy on demand, optimizes loading time, and provides scientific computing capabilities;
2. Chart.js: Draws statistical charts such as histograms and Q-Q plots;
3. Hash routing system: Implements single-page multi-stage navigation via URL hashes;
4. Warm color design: Consistent with the color scheme of paper charts to ensure visual consistency.

The application divides the simulation process into five stages:

• Stage 0: Problem Definition: Guides users to clarify objectives, boundaries, and key performance indicators;
• Stage 1a: Input Modeling: The most complete function, supporting CSV upload or using sample data (stationary exponential distribution/non-stationary peak data), automatically completing data cleaning, distribution fitting (exponential/log-normal/Weibull, etc.), KS test, Q-Q plot visualization, and SimPy code generation;
• Stage 1b: Model Creation: Placeholder, with the goal of converting natural language system logic into SimPy code using large language models;
• Stage 2: Execution: Retains the 2025 emergency room demo archive, and will integrate AI monitoring and adaptive sampling in the future;
• Stage3: Experimental Analysis: Focuses on experimental design and result analysis, with AI applications including automatic identification of key factors and report generation.

Agentic AI Demo: Based on Anthropic's MCP protocol, runs the simpy-mcp server to record LLM sessions, demonstrating AI agents independently running simulation experiments, analyzing results, and iteratively optimizing parameters; Technical Details:

• Pyodide Innovation: WebAssembly compiles the Python interpreter, with an initial load of about 50MB modules, subsequent offline operation, and lazy loading strategy to optimize the experience;
• Pure Static Architecture Advantages: Zero server cost, unlimited scalability, privacy protection (data does not leave the browser), and can be installed as a PWA for offline use.

• Educational Value: A zero-threshold entry platform, no environment configuration required, reducing the simulation learning curve;
• Research Value: Demonstrates the path of AI-enhanced simulation, providing an extensible framework (automatic distribution fitting, agent-based experimental design, etc.);
• Practical Value: The browser-first approach is easy to integrate into existing workflows, no IT approval needed for software installation—just open the browser to model.

Future Plans: Complete Stage1b/2/3 functions, add large language model comparisons, expand distribution types and fitting methods, integrate more simulation engines; Conclusion: This tutorial is an important milestone in the integration of generative AI and simulation. It lowers the entry threshold through browser-side scientific computing, demonstrates the application potential of AI in various stages, and will promote the transformation of simulation modeling from an expert tool to a popularized decision support method in the future.