Much of the world runs on written rules: statutes, building codes, contracts, curricula, licensing requirements. Software that answers questions about them has to be right.
We work on encoding those rules so a machine can check them, and on how much of that work carries from one domain to the next.
Language models handle natural language well and are unreliable at reasoning over many discrete rules. We leave the language to them and give the reasoning to systems whose output can be checked.
We encode written requirements in a formal, machine-readable form. A machine can then verify them instead of paraphrasing them.
Compliance and planning questions have exact answers. A solver computes them and records the steps, so any answer can be inspected afterwards.
A representation built for one domain should work in the next. How much of it carries over, and what it costs to move, is the question we most want to answer.
We take a domain from the first formal encoding to a system people use every day.
AI infrastructure for regulated domains, starting with higher education. Turns university catalogs into degree requirements an engine can check and builds graduation plans that respect prerequisites.
Permitting, professional licensure, tax and contracts are at earlier stages. We write up what we learn for academic venues in AI and law.
Collaborations, engagements, and student projects are welcome.