Gala: Global LLM Agents for Text-to-Model Translation

TL;DR

Gala addresses the challenge of translating natural language descriptions into MiniZinc models by introducing a global-constraint driven, multi-agent framework. Specialized LLM agents target specific global constraints, emitting constraint snippets that an assembler integrates into a coherent model, reducing the cognitive load of single-shot generation. Early results show competitive or superior performance to baseline prompting strategies on Text2Zinc, particularly with stronger LLMs, highlighting the value of problem decomposition and coordinated assembly. This modular approach bridges constraint programming primitives with agent-based modeling, offering a scalable pathway toward reliable NL-to-CP pipelines with practical impact in automated optimization tasks.

Abstract

Natural language descriptions of optimization or satisfaction problems are challenging to translate into correct MiniZinc models, as this process demands both logical reasoning and constraint programming expertise. We introduce Gala, a framework that addresses this challenge with a global agentic approach: multiple specialized large language model (LLM) agents decompose the modeling task by global constraint type. Each agent is dedicated to detecting and generating code for a specific class of global constraint, while a final assembler agent integrates these constraint snippets into a complete MiniZinc model. By dividing the problem into smaller, well-defined sub-tasks, each LLM handles a simpler reasoning challenge, potentially reducing overall complexity. We conduct initial experiments with several LLMs and show better performance against baselines such as one-shot prompting and chain-of-thought prompting. Finally, we outline a comprehensive roadmap for future work, highlighting potential enhancements and directions for improvement.

Figures (1)

• Figure 1: Gala: Global LLM Agents architecture for global-constraint detection and assembly.