Haladir

Jibran Hutchins is a Carnegie Mellon University computer science student who published in IEEE and Elsevier Q1 journals for operations research and machine learning while still in high school, then co-founded Haladir in September 2025 before the ink on his CMU enrollment had dried. The research publication credential is the most important thing to understand about this founding team: these are not students who pivoted into AI after reading about it. They were doing peer-reviewed academic work in the specific technical disciplines (OR, ML, formal methods) that Haladir is now commercialising, before any of them had a college ID. At CMU, Jibran's context deepened considerably. Carnegie Mellon is where operations research was formalised as a modern discipline, where Herbert Simon co-invented AI in the same building where economists and decision theorists worked on bounded rationality, and where the most rigorous graduate programs in robotics, formal verification, and ML engineering in the world still operate. That is not coincidental context. Haladir's core thesis, that formal specification of operational domains unlocks RL generalization in the same way formal verification unlocked AI code generation, is a direct descendant of the CMU intellectual tradition. Before co-founding Haladir, Jibran did software development and private equity research at Grant Park Holdings in New York over the summer of 2025, and prior to that spent four months in PE research at McLean Investment Group in Washington DC. Those two stints are revealing: he was working at the intersection of financial analysis and software development, learning how capital allocates to complex operational businesses, at the same time he was developing the intellectual thesis that would become Haladir. His LinkedIn description of the company, "Haladir helps AI integrate and understand logistics supply chains via OR and ML," is the most compressed version of the product thesis he offers publicly. He posts infrequently and specifically. The YC announcement post named all three co-founders before describing the company. The RLFR technical post came before any product announcement. Both choices are revealing about how he thinks outreach and credibility should work: show the people, show the research, let the product speak later. At YC's W26 batch, Haladir was described as "applied AI product lab for verifiable domains, enabling reinforcement learning and model harnesses to scale to economically complex tasks via the formalization of the informal." The phrase "formalization of the informal" is Jibran's framing. It is precise in a way that most YC pitches are not, and it maps directly to his background: a student who published OR research at research-journal standards in high school, then watched PE capital flow through operationally complex businesses in New York and DC, and decided the missing layer was the formal constraint infrastructure that makes AI actually trustworthy in those environments.

Joseph Tso is a computer science student who was accepted to Princeton University, enrolled in September 2025, spent approximately two months attending class, and then left to co-found Haladir within a week of making the decision, describing the process as "relatively easy." That brevity is not callousness. It reflects a founder who had already made the decision intellectually before he made it logistically, having spent the prior three years doing serious academic research as a part-time researcher at George Mason University's Department of Computer Science from July 2022 to June 2025, a span that runs from approximately his freshman year of high school through graduation. George Mason's CS department has active research groups in operations research, distributed systems, and applied ML. Running research experiments part-time for three years, from high school age, while also publishing in IEEE and Elsevier Q1 journals alongside the rest of the Haladir founding team, means Tso arrived at Princeton not as a student beginning his intellectual formation but as a practitioner looking for the next environment to build in. He found it in San Francisco. Before Princeton, he also interned at Knot, a New York-based fintech startup working on transaction-level data infrastructure, for three months in summer 2025, which gave him direct exposure to the data normalization and enterprise integration challenges that Haladir's Substrate layer is now solving for logistics. His description of Haladir in The Daily Princetonian interview is the clearest public articulation of the product's intellectual core: "operational superintelligence, enhancing the ability of AI to synthesize complex information and determine the best course of action." The specific word choice, synthesize rather than process, determine rather than predict, reveals a founder who thinks about AI as a judgment system rather than a pattern-matching system. He told the Princetonian he feels the company was "something unfinished" at Princeton, and that he intends to return eventually, but that for now the value of being in SF outweighs it. His YC group mentors told him to wear the Princeton dropout status with pride. He does. The combination of George Mason research from early high school, a Princeton CS admission, a Knot internship in fintech data infrastructure, and IEEE/Elsevier Q1 publications before any of this, produces a co-founder who has done more technically rigorous work before age 20 than most engineers do by 30. He is the person on the founding team who articulates the Haladir thesis most precisely in external contexts, and his email, joseph@haladir.com, is the most direct path to someone thinking seriously about the formal constraint layer for RL at operational scale.

Preston Schmittou is currently a freshman at the University of Virginia's College at Wise, a liberal arts institution in rural Appalachia with approximately 1,800 students. He is also a co-founder of a company backed by BoxGroup, Susa Ventures, Y Combinator, and SV Angel, with $4.3M raised, actively serving a leading frontier AI foundation model company as a customer. That gap is the most striking thing about him. UVA Wise is not a traditional feeder for San Francisco AI infrastructure startups. It is a small, access-focused regional college, the kind of institution that educates first-generation college students from Appalachian Virginia rather than placing graduates at top VC-backed companies. Preston's presence there while simultaneously co-founding Haladir is either the product of geographical circumstance or a deliberate choice, and his technical credentials suggest the latter interpretation is more credible. He published in IEEE and Elsevier Q1 journals in operations research and machine learning in high school, alongside the other three Haladir founders, before any of them had enrolled in the institutions they would later leave. In June and July 2024, he did an ASSIP (Aspiring Scientists Summer Internship Program) research placement at George Mason University's CS department, the same research environment where Joseph Tso spent three years, which is not a coincidence and suggests the four founders found each other through shared research context rather than shared school affiliation. From June to August 2024, while still in high school, he also interned at NearStar Fusion, a nuclear fusion startup based in Chantilly, Virginia, working on a technology that sits at the absolute frontier of physics and engineering. Nuclear fusion engineering is not a standard high school internship. It requires comfort with highly constrained physical systems where formal mathematical models govern every parameter, and where correctness cannot be approximated, only achieved. That intellectual environment, where the math must be exactly right because the physical stakes are existential, maps directly onto Haladir's formal constraint thesis: the reason current AI fails in logistics and complex operational environments is the same reason it would fail in fusion reactor control, there is no tolerance for hallucinated answers when real-world constraints are load-bearing. Preston's YC profile bio, "Freshman at UVA Wise. Learning stuff," is either the most deadpan self-description in the W26 batch or a statement of genuine philosophical humility about how much remains to be learned in the work he is actually doing. His email is preston@haladir.com and he is reachable there.

Quan Huynh attends the University of Virginia and is the fourth co-founder of Haladir, the one with the fewest public details available and the most understated self-description: his LinkedIn bio reads "I code sometimes." That phrase, applied to someone who published in IEEE and Elsevier Q1 journals in operations research and machine learning in high school, and who is now co-founding a company working with a leading frontier AI foundation model company on RL training infrastructure for logistics, is not accurate as a professional summary but is accurate as a statement of values: he is a builder who codes when it matters, and does not need the credential to announce the work. The four Haladir founders found each other through shared research context, the George Mason University research environment appears in both Joseph Tso's and Preston Schmittou's backgrounds, and the IEEE/Elsevier publications that all four contributed to before college are the document trail of how they built a shared technical vocabulary before building a company together. Quan's specific technical contribution to the founding team is not publicly documented beyond his co-founder status, but the YC company description is instructive: "data and training infrastructure company for verifiable domains, enabling RL to scale beyond boutique environments toward continuous self-improvement." The phrase "verifiable domains" is Haladir's most technically precise public commitment. Verification, the formal mathematical process of checking that a system behaves exactly as specified, is not a software engineering skill that most founders claim fluency in. It is closer to a research discipline. Publishing in Elsevier Q1 journals for OR and ML from high school requires exactly the kind of formal mathematical reasoning that verification demands. The Haladir founding team is building the infrastructure layer that makes logistics formally verifiable so that RL can scale inside it. That thesis requires four people who can actually do the formal specification work, not just describe it. Quan is one of them. The company is working with a leading foundation model company already, meaning the RL training infrastructure is live with a real customer before most people have heard of Haladir at all. For anyone who works at the intersection of formal methods, operations research, and ML systems, quan@haladir.com is the most direct path to a team doing the kind of work that usually lives only in research papers, now deployed in production for both logistics operators and AI labs simultaneously.