Applied AI Direction · Miami, Florida
A business operator who learned to direct AI fleets, rather than an engineer who learned business. The work is specification, architecture and verification. The code is produced by AI systems under that direction, which is how one person now ships what used to take a team.
InvestorPass is a multi-tenant identity verification platform for private capital markets. It is live, it serves separate tenant deployments, and it exists because the problem it solves was in the way of another business he runs.
He specified it from the domain model through authentication to production deployment, ruled the architecture at every fork, reviewed the work, and tested the result. He did not write the application code, and that distinction is stated here rather than left for an interview to discover.
Thirty-three agents with scoped tool grants and boundaries that are enforced rather than described.
One collector per data source, so no two agents fight over the same material. Read-only gatherers kept structurally separate from the agents that can write. A single named authority per subject, so one question has exactly one answer and two sessions cannot disagree about a fact. Approval gates on anything that alters a record or leaves the building, which means the system drafts and stages but never sends.
Several of those sessions run at once against one codebase without colliding, which is the part most multi-agent setups get wrong.
The distinctive work is not the agents. It is the governance built around them, because an AI system that is right most of the time and silent about the rest is worse than no system at all.
A provenance ladder. Every claim sits on exactly one rung, from a primary artifact held by the business down to an inference with its basis stated. A derived internal document is never a foundation. Anything load-bearing re-anchors on a primary artifact or an external register before it ships.
Indexes generated from their own sources. Each line of the memory index is rendered from the file it points at, so the index cannot drift from what it indexes. Changing a line means changing the file that line describes.
A verifier that reports what no longer resolves. It walks every assertion about where something lives and flags what has gone missing, moved, or grown stale since it was last checked. It runs automatically on every write to a governed file.
A calibration harness that breaks its own safeguards. Each test copies the real signed log into a sandbox, damages it one specific way, and passes only if verification catches the damage. The principle is written into the code: a test that passes by failing to detect its own break is an alarm that does not ring, which is worse than no alarm because it is trusted.
A decision log that reaches the next session. Rulings are captured in the decision-maker's own words and injected into every new agent session before any work starts, which is the fix for the failure that makes multi-session agent work unreliable in practice. 528 rulings recorded, 482 of them on the platform program alone.
Fifteen disciplines, each rated against practitioners applying AI to business operations rather than against AI researchers. The ratings below are published complete, including the two that sit at the floor, because an inventory that claims everything proves nothing.
| Discipline | Rating |
|---|---|
| Context engineering | 5 |
| Grounding and verification design | 5 |
| Decision governance | 5 |
| Specification and work orders | 5 |
| Agent fleet architecture | 4 |
| Multi-session orchestration | 4 |
| Product direction of an AI-built application | 4 |
| Domain corpus construction | 4 |
| Voice and brand as enforceable specification | 4 |
| Integration and automation wiring | 4 |
| Personal data pipelines | 4 |
| Failure analysis and system repair | 4 |
| Security and access control design | 3 |
| Marketing automation and audience growth | 2 |
| Hands-on software engineering | 2 |
Founded a private investor forum and verification platform serving family offices, venture firms and ultra-high-net-worth allocators. Three forums in the first four months, more than 100 vetted investors, combined assets under management above $150 billion.
Joined at concept stage and built the vendor and partner infrastructure that took the platform from zero operational capability to client serviceability in more than 70 countries.
A licensing and operations consultancy scaled to $160,000 per quarter within nine months. 33 concurrent projects as sole consultant, zero errors and zero point deductions across every state scoring submission, and more than 1,500 pages of specifications and operating procedures authored.
More than $400,000 in sales within the first 30 days, tripling the company's standing monthly record, selling training to private military contractors.
Artillery fire direction control, convoy security, marksmanship coach. Two combat deployments to Iraq. Coached more than 1,200 Marines to a 100 percent first-time qualification pass rate.
Every figure on this page was counted from the git history, the session transcripts on disk, or a live page fetched and read. Nothing is estimated, and where the evidence is thin the entry says so rather than rounding upward.