AI Infrastructure & Trust
What makes generative AI worth shipping, not just demo-able. Grounding, factuality measurement, evals, real-time data freshness.
Google Grounding · Factuality Agent · FreshServe · AI Search · Gemini
001 / POSITIONING
Senior Product Manager responsible for Google's AI Search infrastructure — sole PM on a 30-engineer team building the grounding and factuality systems behind AI Search and Gemini. Twenty years shipping consumer products at scale before that: Amazon, Marriott, Walgreens.
A Claude-powered assistant grounded on my career context — responds in first person, in my voice. Useful for sanity-checking whether there's a match before writing a real email. Five questions per session. Ask the good ones.
In May 2024, AI told users to put glue on pizza and eat rocks. Since then, teams like mine at Google have built the infrastructure that prevents that category of failure. Here's what grounding actually does — toggle between the two states.
Grounding is how AI learns to cite its sources. The left side is what a model says from training alone. The right side is what happens when that same model is forced to answer from verified, real-world sources. My team builds the measurement layer that tells us which is which, at scale.
Four projects where I owned the outcome — AI infrastructure at Google, a new-venture consumer launch for Marriott, a loyalty-meets-health platform at Walgreens, and a global marketplace at Amazon. Use the arrows to move through them.
Five additional projects from Google and Publicis Sapient — platform reliability, ML personalization, a national grocery partnership with Google Cloud, a customer data platform, and a top-ranked mobile banking app.
What I think the job of product manager has become — and what's about to become rare enough to matter.
What changes when software is no longer the bottleneck
The cost of turning an idea into working software has collapsed in 18 months. A PM with good judgment and a Claude subscription can ship what used to take a six-engineer team six weeks. I know because I do it.
Here's what actually shifted: six months ago, autonomous agents broke after three minutes. Today they run coding tasks for six hours. That's a phase change. The PM job quietly stopped being about issuing instructions and started being about architecting the systems agents run inside.
Anyone can swing a hammer. Very few can produce finished carpentry. AI puts the tools in everyone's hands; taste is still rare. Taste is still the job.
A mental map of twenty years — four themes, not a list of jobs. Every role I've had sits in one of these buckets.
What makes generative AI worth shipping, not just demo-able. Grounding, factuality measurement, evals, real-time data freshness.
Google Grounding · Factuality Agent · FreshServe · AI Search · Gemini
Taking ML from the lab into products people actually use every day.
Marriott Homes and Villas · Walgreens Mobile App · Albertsons × Google · Marriott Offers · Albertsons CDP · Amazon Appstore · ActiveHealth / Aetna
Shipping for millions when reliability and polish are the whole product.
Amazon Appstore · Amazon Kindle · U.S. Bank · Bank of America · Walgreens Mobile App · Merrill Lynch · ActiveHealth / Aetna
Building the function, not just working inside it — across retail, healthcare, financial services, and telecoms.
Publicis Sapient · KPMG Digital Practice · BearingPoint · Razorfish · Public Sector
Outside of work, I DJ under the name Bigfoot. Twenty years of digging through crates and playing sets around New York. Tell me a vibe and I'll find you something to play.
Recommendations are pulled from my Last.fm scrobble history — twenty years of every song I've actually played. Each result comes with a Spotify preview, and you can save the whole set as a Spotify playlist at the end.