The most valuable resource in a factory is not the steel or the silicon. It is the senior planner who can translate a foreman's complaint about a bottleneck into a linear constraint that a solver can understand. Harumi, a San Francisco-based startup, is betting that a large language model can be that translator [harumi.io, 2024].
Their pitch is a simple swap: instead of a months-long consulting project to build a custom optimization model, you describe your problem in plain English. The LLM reasons through the business logic, writes the mathematical formulation, and generates the code for a solver like Gurobi. It is operations research, but with the friction of modeling supposedly scraped away.
A wedge into the real economy
Harumi is not chasing the usual tech-savvy early adopters. Its stated targets are manufacturers, retailers, and logistics providers, the “real economy” companies where spreadsheets and tribal knowledge still govern asset utilization [harumi.io, 2024]. The wedge is specific: cutting stock optimization. Solve that one painful, quantifiable use case, and you have a foot in the door to tackle scheduling, pricing, and transport.
The company raised $600,000 in a pre-seed round led by Brazilian venture firm ABSeed to build out this premise [harumi.io, 2024]. The team, estimated at 11-20 employees, lists specialties in optimization and industrial engineering [Prospeo, 2026]. While the founders keep a low public profile, the company has garnered recognition in its apparent secondary market, being named one of the most innovative Brazilian startups by KPMG [harumi.io, 2025].
The case for automated modeling
The potential efficiency gains are not trivial. If Harumi’s AI can consistently shave even a single percentage point off material waste for a mid-sized manufacturer, the climate math starts to pencil out quietly. The business case pays for itself; the emissions reduction is a byproduct.
Harumi's early positioning suggests a platform approach, with several optimization surfaces already outlined:
- Cutting stock. The flagship use case, aiming to minimize raw material waste in manufacturing [harumi.io, 2024].
- Pricing optimization. Dynamically setting prices to maximize revenue [harumi.io, 2024].
- Transport management. Routing and scheduling for logistics fleets [Capterra, 2026].
- Crew scheduling. Allocating human resources against shift and skill constraints [Capterra, 2026].
The solver in the room
The most immediate counterfactual is not another startup, but the entrenched incumbent: the operations research consultant. Firms like Gurobi and IBM offer powerful solvers, but they require expert modelers to wield them. The risk for Harumi is that its AI becomes a sophisticated prompt wrapper for these same solvers, never quite capturing the nuanced business context that a human expert would. Trust, in this domain, is built over years, not through a chat interface.
Furthermore, the public traction is opaque. There are no named customers or detailed deployment case studies in the available sources. An estimated annual revenue of ~$1.37 million suggests some commercial activity, but it’s unclear if that comes from a few pilot projects or a growing SaaS base [Prospeo, 2024]. The next twelve months will be about moving from a promising tool to a relied-upon system. That means landing a flagship customer in a hard industry and proving the model’s recommendations are not just clever, but correct and reliable under real factory-floor pressure.