Skip to main content

Technical Architecture

Overview of YuLan-OneSim's system architecture and technology stack.

System architecture​

System Components

Scenario Auto-Construction Subsystem​

  • Function: Converts natural language descriptions into executable simulation code.
  • Key Components:
    • ODD Protocol Translator: Uses the Overview, Design Concepts, and Details (ODD) protocol for formalizing simulation models.
    • Behavior Graph Construction: Models agent behaviors and decision-making logic via a structured graph.
    • Code Generation Engine: Automatically generates simulation code based on user input.
  • User Interface: Allows users to define scenarios using natural language prompts or templates.

Simulation Subsystem​

  • Function: Executes and manages large-scale simulations with real-time monitoring.
  • Architecture:
    • Fully Responsive Agent Framework: Enables dynamic, event-driven interactions between agents and environments.
    • Distributed Master-Worker Architecture: Scales up to 100,000 agents by distributing computational load across multiple nodes while maintaining global consistency.
    • Simulation Execution Engine: Manages agent actions, environment updates, and interaction logic.
  • Features:
    • Real-time visualization of simulation events.
    • Flexible configuration of agent profiles, population sizes, and environmental parameters.

Feedback-Driven Evolving Subsystem​

  • Function: Improves simulation realism through feedback integration (from AI or humans).
  • Framework:
    • VR²T Framework (Verifier–Reasoner–Refiner–Tuner): A multi-agent mechanism that evaluates simulation results, identifies issues, and fine-tunes the LLM backbone.
    • Supports both automated error correction and human-in-the-loop refinement.
  • Goal: Ensures continuous improvement and alignment with expected behavioral patterns.

AI Social Researcher Subsystem​

  • Function: Automates the end-to-end social science research process.
  • Modules:
    • Experiment Design Module: Formulates hypotheses, selects scenarios, and configures simulations.
    • Report Generation Module: Analyzes simulation outputs and generates technical reports in LaTeX format.
  • Workflow:
    1. User inputs a research topic.
    2. The AI researcher formulates questions, designs simulations, runs them, and interprets results.
    3. Generates structured reports with visualizations, analysis, and conclusions.