ORMind is a cognitively-inspired multi-agent framework for solving Operations Research (OR) problems using Large Language Models (LLMs).

ORMind aims to improve LLM-based optimization through:

• A structured, human-like workflow for problem-solving
• Counterfactual reasoning to refine solutions
• Multiple specialized agents working collaboratively

Key components:

• Semantic Encoder
• Formalization Thinking
• Executive Compiler
• Metacognitive Supervisor
• System 2 Reasoner

• Outperforms existing LLM-based OR methods on benchmark datasets
• Reduces compile and runtime errors compared to baselines
• Employs counterfactual analysis to identify and correct errors
• Mimics human expert problem-solving processes

python run_exp.py (for NL4Opt datasets) python run_exp_ComplexOR (for ComplexOR datasets)

• Python 3.7+
• langchain==0.2.7
• langchain-community==0.2.7
• numpy
• tqdm
• gurobipy==10.0.2
• Openai api key

git clone https://github.com/XiaoAI1989/ORMind.git
cd ORMind
pip install -r requirements.txt

ORMind achieves state-of-the-art performance on:

• NL4Opt dataset: 68.8% accuracy
• ComplexOR dataset: 40.5% accuracy

If you use ORMind in your research, please cite our paper:

ORMind: A Cognitive-Inspired End-to-End Reasoning Framework for Operations Research

This paper has been accepted by the Industry Track of the 63rd Annual Meeting of the Association for Computational Linguistics (ACL 2025)