About
Twenty years of building.
Now at full speed.
PhD neuroscientist → venture-backed founder → product leader at some of the biggest tech companies in the world → solo AI engineer. Every stop taught me one thing: the distance between "we have the data" and "the team can use it" is where value dies. I close that distance.
The short version
Scientist first
PhD in behavioral neuroscience; postdoc at Columbia, Weill Cornell, and Rockefeller. Seven peer-reviewed publications. I still work like a scientist: instrument everything, validate honestly, publish the negative result.
Founder twice
Built Stirplate, a "GitHub for science data," to paying customers at Cornell, Mt. Sinai, and Harvard on $800K raised. Now building Hidden Facets and LucenArc.
Product at scale
I've worked at some of the biggest tech companies in the world: I took a new cloud product from a single site to 21 countries with billions in revenue impact, served as Director of AI product at a multibillion-dollar company, and owned product lines worth 20% of a cloud provider's revenue.
Builder again
The past year: roughly twenty production AI systems shipped solo, spanning sales intelligence, automated QA, behavioral analysis, knowledge retrieval, and cross-system data architecture.
A translator between worlds
This is the part that's rare. I'm not only the engineer who builds it. Fifteen-plus years as a product manager at some of the world's biggest tech companies, plus two founder runs of my own, means I speak business, product, data engineering, and ML research fluently. I can sit with your executives, your operators, your data engineers, and your ML researchers and translate between all of them, down to the nitty-gritty of what you actually need. Most AI work fails there, not on the model, but because no one in the room understood the problem from every side. That's the seat I take.
How I keep AI honest
Principles from production, not a slide deck.
Code computes, AI formats
Any number a human sees is computed deterministically by Python. The model only turns it into a sentence. No hallucinated metrics, ever.
Read-only by default
Your systems are integrated with read-only credentials unless we explicitly agree otherwise. I connect to your stack; I don't mutate it.
Verified before visible
Trust tests that challenge correct answers with false information. Cross-agent verification that re-queries source systems. Invariant checks that block known-bad outputs before they reach users.
Discovery before code
Every integration starts with full schema discovery: what fields, objects, and endpoints actually exist. We build on what's there, not what we assume.
Bring me one problem.
If I can't solve it fast, I'll tell you before you spend a dollar. The first build usually speaks for itself.
Start a project →