Optimization for people who will never write a MILP.
Warehouse and category managers at industrial companies typically work by reading ERP screens and trusting that the numbers are correct. They have no background in statistics or optimization — and cannot sanity-check a wrong answer the way an operations-research specialist could.
How it works
Language in, verified model out. Inspired in approach by Stanford’s OptiMUS: structured extraction, a per-constraint connection graph, confidence-based human escalation and an iterative debug loop.
Constrained by design. Instead of free-form model generation, the pipeline maps requests onto a library of pre-verified templates for known supply-chain problem types.
Explainability as a module. Explanation is a first-class component alongside ingestion, forecasting and optimization — recommendations arrive with the reasoning a non-specialist needs.
Under the hood
References
- Stanford OptiMUS — arXiv:2407.19633 (approach reference, no shared code).