Sailor helps fix missing numbers in optimization problems interactively
SAILOR: Solver-Assisted Interactive LLM-based Optimization Recovery
Artificial Intelligence
Summary
Optimization problems need exact numbers to find the best solutions, but sometimes descriptions are missing those numbers. The authors created SAILOR, a system that spots missing key numbers, asks specific questions to get them, and then updates the problem to find better solutions. They tested it on many examples using simulated answers, showing it can improve solution accuracy with only a few questions. This approach is a first step toward smarter tools that can help users define optimization problems more clearly.
What this means in practice
- •For operations teams: Clarify missing numerical details in supply chain optimization problems by asking targeted questions during problem setup.
- •For financial planners: Improve budget allocation models by interactively recovering missing cost and demand values for more accurate solutions.
Tested on simulated data.
Authors
Shaghayegh Sadeghi, Stephen L. Smith, David C. Del Rey Fern'andez
Abstract
Natural-language descriptions of optimization problems may be incomplete or vague about numerical information that a solver requires, including costs, capacities, demands, bounds, and penalties. A language model can translate the description into code, but when a required value is absent it must either stop or guess. We present SAILOR, a proof-of-concept system that detects such unsupported numerical choices, asks the user targeted follow-up questions, and updates the optimization model before returning a solution. Questions are prioritized using uncertainty and solver-derived estimates of how strongly each missing value affects the current model. We evaluate the pipeline on 1,723 instances from seven masked benchmarks using an idealized simulator that returns ground-truth values. Exact objective-value agreement ranges from 27.0% to 87.6% across datasets, with 1.4--5.7 questions per instance on average. These results establish feasibility under controlled branch-and-reveal feedback; they do not measure performance with human users or general structural model repair. Code is available at: https://github.com/sshaghayeghs/SAILOR.