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

TL;DR

ORMind presents a cognitive-inspired end-to-end reasoning framework for operations research that replaces brittle multi-agent orchestration with a structured, memory-coordinated workflow guided by counterfactual reasoning. It maps business requirements into mathematical formulations using a Semantic Encoder and Formalization Thinking, then translates them into executable solver code via an Executive Compiler, System 2 Reasoner, and Metacognitive Supervisor within a Memory Pool. On NL4Opt and ComplexOR benchmarks, ORMind demonstrates strong problem-formulation accuracy and competitive enterprise performance, with practical validation from Lenovo’s AI Assistant deployment, highlighting improvements in reliability, transparency, and efficiency. Ablation and robustness analyses underscore the critical roles of modular components and deterministic prompting, suggesting a scalable, explainable approach for industrial decision support. The work lays groundwork for cognitively aligned AI architectures in enterprise optimization beyond OR, with future work focusing on larger datasets and optimized resource coordination.

Abstract

Operations research (OR) is widely deployed to solve critical decision-making problems with complex objectives and constraints, impacting manufacturing, logistics, finance, and healthcare outcomes. While Large Language Models (LLMs) have shown promising results in various domains, their practical application in industry-relevant operations research (OR) problems presents significant challenges and opportunities. Preliminary industrial applications of LLMs for operations research face two critical deployment challenges: 1) Self-correction focuses on code syntax rather than mathematical accuracy, causing costly errors; 2) Complex expert selection creates unpredictable workflows that reduce transparency and increase maintenance costs, making them impractical for time-sensitive business applications. To address these business limitations, we introduce ORMind, a cognitive-inspired framework that enhances optimization through counterfactual reasoning. Our approach emulates human cognition, implementing an end-to-end workflow that systematically transforms requirements into mathematical models and executable solver code. It is currently being tested internally in Lenovo's AI Assistant, with plans to enhance optimization capabilities for both business and consumer customers. Experiments demonstrate that ORMind outperforms existing methods, achieving a 9.5\% improvement on the NL4Opt dataset and a 14.6\% improvement on the ComplexOR dataset.

Figures (3)

• Figure 1: Current frameworks rely on complex agent orchestration with unpredictable execution paths, dramatically increasing API calls and computation time. Their focus on code syntax rather than mathematical accuracy results in costly errors that can propagate through business operations undetected. This excessive coordination overhead makes these systems impractical for time-sensitive business applications. Compared to traditional methods, ORMind employs a streamlined end-to-end workflow with counterfactual reasoning, significantly enhancing solution reliability.
• Figure 2: Our approach is grounded in established cognitive science theories, particularly dual-process frameworkkahneman2011thinking and tripartite model of cognitionstanovich2009distinguishing. The Semantic Encoder and Formalization Thinking modules correspond to Type 1 (intuitive) processing, while the System 2 Reasoner implements Type 2 (analytical) processing. The Metacognitive Supervisor embodies the reflective mind, monitoring and coordinating between these systems.
• Figure 3: Temperature analysis on NL4Opt and ComplexOR