Frontier models can already do graduate-level synthesis planning, novel-agent design, and supplier search. What's missing is the layer that ensures they never cross a line — export control, select agent, controlled substance, dual-use chemical. We build that layer.
Defense, biosecurity, and chemistry are not domains where models can move fast and break things. They are domains where a single uncaught synthesis is a violation, an incident, or a casualty.
Night Shepherd builds the safety infrastructure that makes autonomous research deployable in those environments — neurosymbolic, auditable, and grounded in statute. We start where output filters end.
Identifying details withheld pending public-release review.
Full-stack and AI engineer specializing in neurosymbolic systems and regulated software infrastructure. Built LIGHTNING at the SCSP AI+Expo Hackathon 2026 — a neural→symbolic→hybrid safety layer spanning 18 export-control regimes with a hard trust property: the LLM never makes the final classification. Prior work spans legal compliance platforms, autonomous racing hardware, and production AI pipelines.
Aerospace and applied-AI engineer. Flight control and guidance systems at NASA Johnson Space Center; inertial navigation systems at Draper Laboratory; computer vision and AI at Argonne National Laboratory. M.Eng., mechanical engineering & manufacturing, MIT. B.S., mechanical engineering, Columbia — team captain, FSAE.
We don't build research agents.
We make them deployable.
The intelligence that wins doesn't live outside regulation. It lives next to it, speaks its language, and shows its work.
For commercial access, regulatory partnerships, or research collaboration.
Open contact form