Sketch2DES: Open-Weight LLM Workflow Converts Queue Diagrams into Simulation Models
Researchers propose Sketch2DES, a sketch-to-simulation workflow that uses open-weight large language models (LLMs) to convert diagrammatic representations of queuing networks into verifiable discrete-event simulation models. The workflow consists of three stages: (1) a multimodal LLM translates a diagram into a semi-structured textual description; (2) an LLM with a reflection-based verification loop converts that description into schema-validated JSON data; (3) a software adapter deterministically transforms the structured data into an executable simulation model. Intermediate artifacts can be inspected and automatically validated before execution, addressing verification and reproducibility challenges for users without programming expertise. The approach is detailed in arXiv:2607.24259.
Key facts
- Sketch2DES converts diagrammatic representations of queuing networks into verifiable discrete-event simulation models.
- The workflow uses open-weight LLMs.
- Stage 1: multimodal LLM translates a diagram into a semi-structured textual description.
- Stage 2: LLM with reflection-based verification loop converts description into schema-validated JSON.
- Stage 3: software adapter transforms JSON into executable simulation model.
- Intermediate artifacts can be inspected and automatically validated before execution.
- The approach addresses verification and reproducibility for non-programmers.
- Paper published on arXiv with ID 2607.24259.
Entities
Institutions
- arXiv