On paper, the pitch sounds almost quaint: marry the rigid certainty of operations research with the fluid reasoning of large language models. In practice, what Haladir is attempting represents one of the trickier technical challenges in the current AI boom—getting models that excel at conversation to reliably handle the hard constraints of real-world logistics.
The San Francisco startup, which emerged from Y Combinator with backing from Susa Ventures, SV Angel, and DoNotPay founder Joshua Browder, bills itself as building "operational superintelligence" for global supply chains. The funding amount hasn't been disclosed. The team is four people, according to the company's YC profile, though LinkedIn lists the headcount somewhere between two and ten employees. And the problem they're attacking has defeated bigger players with deeper pockets.
Two Worlds, One Platform
Most logistics software lives in one of two camps. Traditional optimization tools—the kind built on mixed-integer linear programming, satisfiability solvers, and constraint satisfaction—can guarantee mathematically optimal solutions. Load this truck first, then that one. Route these drivers exactly this way. The math checks out every time. But introduce ambiguity—an unclear customer request, a vague delivery instruction—and these systems choke.
LLMs, by contrast, handle ambiguity beautifully. They parse messy human language, infer intent, navigate context. What they can't do, at least not reliably, is enforce the ironclad rules that logistics operations demand. You can't put 40,000 pounds in a truck rated for 35,000, no matter how persuasively the model argues otherwise.
Haladir's founders—CEO Jibran Hutchins from Carnegie Mellon, along with Quan Huynh, Preston Schmittou, and Joseph Tso, all from Princeton computer science—are building a platform that attempts to bridge that divide. Whether they can actually pull it off is another matter.
The architecture has three layers. "Substrate" ingests data from the alphabet soup of logistics systems: warehouse management, transportation management, yard management, order management. It converts all of that into what the company describes as a "queryable operational graph." "Operator" deploys AI agents with real execution authority, handling dock assignments, picking, packing, exception handling—all within policy guardrails. "Engine" runs the heavy optimization: vehicle routing, multi-echelon inventory, demand forecasting, labor scheduling.
Their target customers are third-party logistics providers and distributors, though the company's YC profile mentions work with "one of the leading foundation model companies." That suggests Haladir may be playing a dual game: selling to enterprise logistics buyers while also providing training data or reinforcement learning environments to AI labs. No customer names have been made public.
Benchmarking the Gap
Earlier this year, Haladir's research team released ConstraintBench, a benchmark spanning ten operations research domains designed to test whether LLMs can handle optimization problems requiring both feasibility and optimality. The results weren't encouraging. State-of-the-art models scored below 31% on tasks that demanded constraint satisfaction and optimal solutions.
The company followed that up with work on RLFR—Reinforcement Learning from Formally-Defined Rewards—aimed at improving code generation. Both efforts position Haladir as something more than a vertical software vendor. They're building infrastructure, publishing research, courting AI labs. It's an unusual posture for a seed-stage company with a handful of employees, though perhaps less so in an era where the lines between enterprise SaaS and AI infrastructure have blurred considerably.
A Familiar Pattern
Haladir isn't alone in chasing logistics automation. BackOps AI raised $6 million for logistics automation in mid-2025. UK-based Magentic pulled in €4.6 million in July 2025 for AI agents in supply chains. Susa Ventures, one of Haladir's backers, has form here—its portfolio includes Flexport and Stord—and closed a $175 million fund recently.
What's less clear is whether a four-person team can execute on a vision that spans data integration, agentic execution, formal optimization, and foundation model partnerships. The technical narrative is ambitious. The backing is credible. But logistics is a grinding, detail-heavy business that doesn't forgive theoretical elegance.
For now, Haladir has the capital and the research pedigree. What it needs next is proof that the math works outside the lab—and that customers will pay for it. That part, as any operations researcher will tell you, is a harder problem to solve.
