I was born in Japan and raised in New York. Before I ever opened an engineering textbook I spent about fifteen years on a pre-college conservatory track in classical piano, which turned out to be the best early lesson I ever got in how most work actually improves. You do it carefully, for longer than feels reasonable, and then you do it again. That's basically the whole thing.
I studied chemical and biological engineering at Princeton and spent my senior year in Dan Steingart's lab pushing on zinc-bromine batteries. We got a paper into Energy & Environmental Science and filed a patent out of it, and somewhere in that process I figured out what I actually like doing: working a consequential problem until the argument is tight enough that someone else can pick it apart on the specifics.
Out of college I went to 24M Technologies as technical lead on a $3.5M ARPA-E battery program. It was the first place I had to treat a result as unreal until it had survived a budget review and a manufacturing process, not just a lab bench. That experience ruined me, in a useful way, for the kind of science that stops at the beaker.
I started Jumpstart Energy in 2018. The original idea was a marketplace that would move clean-energy capital into places where new infrastructure actually mattered, which was a very appealing thesis that mostly did not work. What the market kept asking for was better underwriting, so I narrowed the company down to a machine-learning credit product for solar loans, bootstrapped it, and sold it in 2019. Jumpstart was the first company I built that had to change shape under me because reality refused to cooperate with the original plan. I learned more from that than from most of the things that went right.
After Jumpstart I spent a year at Mission Lane as a senior data scientist on fraud and credit decisioning. It's hard to overstate how different production ML inside a consumer bank feels from anything that looks like research. Every model I touched was a live set of decisions with real people on the other side of a threshold I'd set the night before.
Philip Seifi and I started Colabra in 2021. We began in pharma and scientific R&D, then followed the market into buy-side M&A diligence, which is where the company lives today. We help deal teams turn entire data rooms into clause-cited issues lists and a clearer picture of the risks on the other side. It took me a while to see it this way, but Colabra is the place where a lot of what I'd been practicing separately up to that point finally had to work together at the same time.
If there is one thing I think I am genuinely good at, it is taking something ambiguous and making it legible enough for someone else to act on. In the lab, that looked like writing a paper clearly enough for another researcher to disagree with me on the specifics. At Jumpstart, that meant turning a broad thesis into a narrow, testable wedge. At Colabra, it shows up every day in how I sell: I figure out where a buyer's workflow actually breaks, I name that moment precisely, and I find the smallest credible next step that gets both of us moving. In practice that means grounding outreach in the buyer's own language, writing different stories for corporate development teams, private equity firms, and advisors instead of pushing one generic deck, and using low-friction on-ramps like a historical backtest or sample issues list instead of asking someone to redesign a workflow on faith. An advisor I worked with told me my cold outreach was a superpower. I think the real thing is just that I take the time to understand the problem well enough that the outreach does not feel cold.
Outside work I run a lot. I'm a 2:58 marathoner, still chasing the next block. I read, I still sit at the piano sometimes, and mostly I still think about craft the way I first learned it there.
