A great deal of the world runs on written rules: statutes, building codes, contracts, curricula, licensing requirements. Software that answers questions about them has to be right, not merely convincing.
We work on making those rules machine-checkable, and on finding out how far the same approach carries from one domain to the next.
Language models are good at language and unreliable at discrete reasoning. Rather than ask them to be something they are not, we give the reasoning to systems that can be checked.
Written requirements are encoded into formal, machine-readable representations. What was prose becomes something a machine can verify rather than paraphrase.
Compliance and planning are discrete problems with exact answers. Solvers produce them, with the derivation available for inspection afterwards.
A representation built for one domain should carry to the next. How far it carries, and what it costs to move, is the question we care most about.
We take a domain from its first formalization through to a system people rely on daily.
AI infrastructure for regulated domains, starting with higher education. Compiles university catalogs into machine-checkable degree requirements and generates prerequisite-aware graduation plans for students.
Permitting, professional licensure, tax and contracts are at earlier stages. Findings are published at academic venues in AI and law.
Collaborations, engagements, and student projects are welcome.