Skip to content

About Jaxon

A Department of Defense problem, turned into AI you can prove

Jaxon was founded by Scott Cohen and Greg Harman out of their work with the Department of Defense. The team they built pairs machine learning engineers with people who have spent careers defending decisions to regulators, examiners, and inspectors general. That is why DSAIL enforces policy the way the people who own the policy actually apply it.

Jaxon, in one minute Watch

A short introduction to what Jaxon builds and why. Watch on YouTube

Our story

Built out of a real problem, not a pitch deck

Every DoD workflow Scott and Greg looked at had the same structural obstacle. Prompt tuning could lower a model's error rate, but it could never produce evidence that a particular output had followed the rules, and in adjudication and classification work that evidence is the whole job.

That work has since been funded through a Space Force Phase III SBIR, an Air Force Phase II SBIR, and Air Force TACFI, and Jaxon is awardable through Tradewinds.

The same requirement turned up in regulated commercial work, which is why decision-makers in financial services, insurance, and life sciences come to Jaxon: CTOs, data science leaders, and analytics groups who cannot ship an AI decision they are unable to stand behind.

How we work

Collaboration first, roadmap second

Our culture

Collaboration is the cornerstone of how we work. We take a team-first approach, and our best ideas come out of working sessions with customers, not a whiteboard in isolation.

Looking ahead

We're expanding our product line and growing the team behind it, while holding to one measure of progress: does this make AI projects succeed more often, and does it get them there faster.

Advisory board

Advisors who have run this at scale before

Former agency directors, military leadership, and industry operators who help us build Jaxon for the way their fields actually work.

Get in contact

Want to see DSAIL® on your own documents?

Tell us the workflow where a wrong AI answer is unacceptable, and we'll show you what a verified verdict looks like on it.