AI PRODUCT MANAGER — AGENTIC SYSTEMS

I build AI you can trust enough to act on.

AI produces numbers far faster than it produces reasons to trust them. When an investor acts on those numbers, that gap costs real money. I built an agentic system that closes it: gated where it matters, graded on every run.

THE DECK

Would you bet real money on an AI’s answer?

Stock-Max is an agentic equity-analysis system I built and operate. The deck walks through the product calls that made it worth the bet, with a receipt behind every claim.

OPEN THE DECK PDF · 11 slides

The case studies proving the output, governing the agent Product judgment

Case study · PDF

Proving the Math, Not Trusting It

How I made an AI’s financial analysis trustworthy enough to act on

I gave the AI’s math a check it can’t influence: the AI builds the analysis as a working Excel model, every number a live formula, and separate Python recomputes each number from the raw inputs. The two have to tie out; they share nothing but the data, so they can’t agree on a wrong answer. Then I planted bad math to prove the check fails exactly when it should.

READ THE STUDY

Case study · PDF

Making the Right Rule Hard to Ignore

How I made an AI agent more reliable by controlling what it knows, and when

An agent can ignore any rule it holds. My job was to make the one that matters the hardest to ignore: a thin always-on core, playbooks that load at the moment of need, and a hard gate that holds a high-stakes run until its rules are in. I made the agent more reliable as its job grew, on 56 percent less context per run.

READ THE STUDY

Operating record

The technical story

analysis runs logged

160+

stock analyses

wrong numbers reached the investor

how this is counted

eval-log findings fixed

The full operating record is in the deck · as of July 9, 2026

About

I built my career shipping financial-crime and KYC product for global banks. Every number I shipped had to survive an audit. I hold AI to the same bar: a system that hands you a number owes you the proof.

I scoped agentic AI strategy for Goldman Sachs’ KYC program, then built and operate my own agentic system end to end, to know what that strategy feels like in practice.

The product calls an AI team needs made: I’ve made each one myself, so I know what I’m asking of the people who build it.

Where the human sits,
what gets gated,
how trust gets proven.

FORMAL TRAINING · agentic AI systems · AI evaluation · ML fundamentals (Stanford / DeepLearning.AI) · full credentials ↗

Off hours: meditation, pickleball, new restaurants with my wife, and our Cavapoo, who somehow has more energy than I do.