Forward-deployed engineer.

AI projects

• Warrior PathA live app that helps Navy SEAL candidates train, and a roster that tells their mentors who is actually doing the work.486 candidates · 145 contracted · 309 shipped to the Navy20 staff, 20 divisions, 2 regions liveMy region: dead last to #2 in the nation (a Navy recruiting command record); FY26 goal ~6 weeks earlyMoreLess

  I mentor for Naval Special Warfare across several states — hundreds of candidates, one of me, and a yearly quota. The only system we had was a database of names. It could tell me a candidate existed; it could not tell me whether he had trained a single day. So I would drive four hours to sit with a kid who worked out because I was standing next to him, while a kid who trained every morning went unnoticed. Effort was invisible.

  I built candidates an app they use on their phone to log training, and mentors a roster that turns that into a measure of effort over time: who is engaged, who is improving, who has gone quiet. It also let me show the Navy which candidates a mentor could genuinely help versus which had never started and needed a different conversation with their recruiter.

  Candidates who engage in their first week sign at 20% versus 1.6% for those who do not. I use that to decide where to spend a Saturday. It is a strong signal, not proof that pushing engagement up causes contracts, and I do not claim otherwise.

• Brief MeA voice agent that has read the entire system it lives in and can act inside it. An hour of dashboards became a fifteen-minute phone call.60 minutes a day → 15Reads the live codebase and dataTakes action, with consent gatesMoreLess

  The usual approach is a chatbot with a paragraph describing your data. It sounds confident and cannot answer a real question. Instead I gave the agent tools to read the actual code and data it lives inside, plus permission to act — message a candidate, email a recruiter, pull one person's history.

  Now I get briefed on my whole region in a conversation I can have while driving, and I can tell it to send the message rather than going back to a screen. The voice agent on this page is the same pattern, pointed at my own record.

• RidgelineThe safety system that decides whether an AI's work is allowed to ship. Every plan gets attacked by other models before anything is built, and only my phone can override it. Now the delivery platform at Uncommon Waters.2,004 pull requests governedAdversarial review before build and before mergeHuman approval cryptographically signedMoreLess

  AI coding agents fail in two ways that matter near production. They are confident and wrong, and once they commit to a path they keep confirming their own decision instead of questioning it. Separately, a shared credential cannot prove a human approved anything — an agent holding the token looks exactly like me holding it.

  So before an agent writes code, its plan goes to a panel of different models whose only job is to break it: find the gap in the reasoning, the thing that cannot be validated, the requirement it quietly dropped. Only an approved plan gets built. The finished code gets a second adversarial pass. Disagreement goes to a separate model to rule on; when even that is not confident, the decision comes to my phone, signed, so the approval is provably human. The only thing that can hard-stop a merge is an automated check — models advise, I judge.

  The reason I trust it is that it works on itself. It once blocked a defect that was inside its own implementing code.

  It is also why I can say this plainly: I do not hand-write code. Every line of these systems was written by AI agents inside this pipeline. I own the architecture, the acceptance criteria, and the audit trail.

• MedFlowAI-run technical due diligence on a healthcare software company — delivered as an audit trail a lawyer can check line by line, not a model's opinion.Every claim independently verifiablePatient data never left our environmentFindings had to survive attack, not agreementMoreLess

  MedFlow is a healthcare and clinical-lab software company. The question was whether its technology and compliance posture were what they had been represented to be. Patient data was involved, so being wrong is not an inconvenience.

  Hand a model a codebase and ask what it thinks and you get fluent agreement and no evidence. I ran it as an evidence process instead: reviews went out as small focused packets rather than the whole repository, the standard for a finding was whether it survived being attacked rather than how many models agreed, and every load-bearing claim carried a check someone else could run. Regulated data never went to an outside model. When one of my own agents reported work it had not done, I removed it and kept the parts of its output that were real.

  I do not publish conclusions about that company's business, and I will not discuss them in an interview.

Advisory

• Defense tech — advisor to the CEOHeavy-lift drone company, pre-seed to acquisition: roughly $1.4 million raised, acquired for about $30 million.MoreLess

  About eighteen months advising the founder on strategy, go-to-market, product, and ultimately the exit. He was twenty and had built a drone you could sit inside. The work was turning that into something the defense innovation ecosystem would treat as a product worth buying — which partnerships to pursue, which people to hire, and how a very young founder holds a room.

• Signal Messenger — Office of the PresidentBuilt the operating engine behind the President's global engagement — email operations, the speaking pipeline, and strategy for interactions with heads of state.MoreLess

  The problem was volume and stakes at the same time. Inbound from world leaders and elite business networks was arriving faster than the organization could triage it, and communications from heads of state were being missed. Deciding between two invitations is easy when one is unimportant; it is hard when both matter.

  I built one place where all of it landed, a single weekly decision forum, campaigns to keep high-value invitations warm rather than declining them outright, and a way for leaders beneath the President to take the engagements that belonged at their level.

• Lineage Logistics — advisor to data leadershipAdvised the Head of Data and leadership on operating systems and data-driven scale through the period leading to the company's IPO.MoreLess

  One of the largest cold-chain operators in the world, scaling into a public offering. The work was the architecture of how decisions get made with data, not a dashboard.

• Aleo — advisor to the CEOAdvised the CEO of the privacy-focused Layer-1 blockchain company.MoreLess

  Strategy and operating counsel for a frontier cryptography company. Part of my consultancy years — about 16 clients between 2022 and 2024, before I closed that chapter to build with AI full time.

• Other engagementsAcross my career: about 60 companies in 20 industries, and about 20 founders coached.MoreLess

  Mostly the same job: design the operating system — the actual architecture and logic of how a company works — and run breakthrough product or innovation programs when the conventional approach has stalled.

  One example of the pattern: a veteran-employment platform with obvious social value and no business underneath it. We split the pitch into employer ROI through workforce tax credits and veteran benefit as the byproduct, and the founder raised on it.

Operating leadership

• SEAL Future Foundation — COO, 2020–2022Built an end-to-end community application that united a fragmented SEAL network and is now the largest population of Navy SEALs online anywhere.~2,400 users~30 sub-communities~60% daily activeMoreLess

  I built the app from scratch. A famously tight community had no single place to ask for help or hire each other after service; now it does, and it absorbs the requests that used to require a staff member — which is what let a tiny team scale.

  I left Geotab, and equity, to take the operations seat when the organization had six weeks of cash. The president could raise money and was drowning in operations, so I took all of it off him and pushed him out to fundraise, then decentralized into local chapters — about four when I arrived, 26 when I left. I hired the CEO, the business development team, and the technical team, and the organization posted consecutive record years.

• Geotab — VP of Operations, 2018–2020Ran ten departments and led the U.S. government relationship for a ~$500M federal / GSA award. Built Customer Success and Retention while the organization scaled from roughly 200 people to 1,500.~5,000 customers onboarded/month~0.02% monthly churnOrg ~200 → ~1,500MoreLess

  I built the customer success and retention function before it was a standard job title, and automated onboarding well enough to keep pace with a company growing that fast — including launching oil and gas in 13 countries and 13 languages at once.

  I also led the U.S. government relationship for a roughly $500 million federal and GSA award, standing that side of the business up from nothing through FedRAMP certification.

  Geotab is where I was introduced to AI in 2018, and where I first put it into a scaled hardware and SaaS operation rather than treating it as an experiment. Building with it became the job in 2024.

• The Scienomics Group — VP of Operations, 2014–2018In about six weeks, moved an account from losing the business to a ~$20M renewal. Built an adaptive physician-education product from scratch and sold it to Amgen for $1M.~$20M renewal$1M product sale to Amgen~35 jobs savedMoreLess

  The team knew exactly which service levels they were missing and tracked them religiously. Nobody had asked what actually produces a service level. I found the handful of daily conditions that determine the outcome, reported on those three times a day, and made the day's list the thing you finish before you go home.

  Underneath that was physician education at real scale: roughly 10,000 events a year supporting about 30,000 physicians.

  I also built a product there from the ground up: an early prototype of adaptive physician education — ask a question, and the content journey changes to match what you know. I sold it to Amgen for $1 million. A decade later, large language models made that concept native. I prototyped and sold it before the technology caught up.

• Wynn and Encore Las Vegas — Senior Assistant Director of Operations, 2012–2014Hired directly by Steve Wynn to transform a 550-person, 24/7 operation. Built eight training curriculums across four employee levels to accelerate competency without increasing the budget.MoreLess

  Two of the largest resorts on the Strip, running continuously, with heavy foot traffic and real regulatory exposure. The mandate was transformation, not maintenance, and I could not spend more to get it.

  So instead of hiring my way out, I built the progression itself: eight curriculums mapped to four levels of employee, so people advanced on demonstrated competency rather than tenure. It is the same instinct behind Warrior Path — make capability visible and give people a path to earn the next thing.

• U.S. Navy SEAL — SEAL Team One, 2003–2010Combat medic. Direct action, special reconnaissance, and hostage rescue.MoreLess

  Where I learned to plan when the consequences are real and to lead people who are tired and afraid. It is also why the Navy later put me in the mentor seat that Warrior Path came out of.

Education

MBA, Drexel University. BA, National University. Harvard Business School professional leadership development program, roughly nine months.

Check my work

The product is live and you can open it right now — warriorpath.ai. Everything else sits in private repositories because it holds candidate or client data, and I am happy to share those or walk you through any project on request.