
Launch Video: https://www.youtube.com/watch?v=caPPFAmL8Iw
Founded by Jibran Hutchins, Quan Huynh, Preston Schmittou & Joseph Tso
In high school, they published and were cited in IEEE and Elsevier Q1 journals for operations research and machine learning. They left Carnegie Mellon, Princeton, and UVA to build Haladir.
Operationally-complex companies (logistics, supply chain, etc.) need to make thousands of decisions a day under tight, real-world constraints. Even when given ample context, they observe that the top models often fail to understand these constraints when placed in difficult scenarios, providing suboptimal or confidently incorrect responses. Though they excel at individual tasks like coding and data analysis, optimizing across systems with millions of interdependent variables under hard constraints remains beyond what current models can do on their own.
Haladir combines formal solvers with LLMs to give models a fundamental understanding of constraints rather than just semantic context. For logistics and supply chain companies, this means routing, scheduling, and resource allocation that AI can reason through rather than guess at.
They build at both the model training layer and the application layer. On the training side, they build solver-based RL environments and data pipelines for frontier labs. On the application side, they build model harnesses that allow deployed agents to operate within constrained, sensitive workflows and make optimal decisions. Decades of development in solvers like Gurobi and SAT/SMT have produced incredibly powerful technologies, but there's a growing gap between these tools and how AI is deployed today. They are here to close that gap.
Read their research: haladir.com/research
Their Ask
The team would love to talk with you if: