How We Use AI in Our Armor Development
Every plate we build starts with a question: will this combination of materials stop the threat, and how much deformation will it leave behind?
For most of this industry, the answer comes from experience and a fresh test. We have been doing this for a while now, and we had a third option sitting in a filing cabinet: our own test history.
The problem with a filing cabinet
Since 2023 we have put a lot of plates in front of a gun at independent NIJ-accredited laboratories. Each one produced a formal report: velocities, penetration results, back-face deformation, clay calibration, conditioning.
That is a real body of evidence. It was also close to useless day to day, because it lived as a stack of PDFs. If you wanted to know how a particular backer thickness performed against a particular threat three years ago, you had to remember which test it was and go find the paper.
So we built a database, and we used AI to do it.
What we actually built
We call it BALLISTIQ. It is an internal engineering tool, not a product.
Every certified lab report we hold has been read, parsed, and turned into structured data: one row per shot. 100+ test panels. Over three hundred individual shots. Four independent laboratories. Every value traceable back to the report it came from.
On top of that sits the part that matters most: the material record. What ceramic, what purity, what thickness. What backer fiber, what grade, how many plies. What foam, what adhesive, what curvature. Because that is what actually carries forward. Model names get retired. Materials do not.
Now a question that used to take an afternoon of digging takes a few seconds, and the answer comes with the specific historical tests it is reasoning from.
The rules we gave it
An engineering tool that tells you what you want to hear is worse than no tool at all. So we were strict about a few things.
The lab report always wins. Where our own notes disagree with a certified test report, the report is right and the note is wrong. We found several of these while building the database. We corrected the notes, not the reports.
A projection is never a result. The tool can estimate. It has to say so. Nothing estimated is ever presented as something that was measured.
Why this matters for what you buy
It means our design decisions are anchored to our own test evidence rather than to a supplier's datasheet or a rule of thumb. When we change a material, we can see what that change did the last time we made it, at what thickness, against what threat.
It also means we know exactly what we have not tested yet, which is the more useful half of the picture.