Product leader and three-time founder. I work in industries where being wrong is expensive. These are the problems I've spent the last decade on.
Have it emit validateable SQL, not answers. A query is something a human can read before it runs, and re-run against any data snapshot afterward. Every deployment was tracked against a fuzzed dataset built from real studies.
studyOS, 2022–2025. Started as analytics for biotechs; discovery showed that was a commodity, so we pivoted to self-service AI reporting for pharma, which is what won the enterprise accounts. Customers inclued Bausch and GSK. Ran it across 8+ studies, 3 of them registrational, on a $3.5M seed. Acquired.
You build the logistics yourself, because nobody sells them. Contracted networks supply phlebotomist contact details and nothing else — no dispatch, no routing, no chain of custody. All of that had to exist before a single draw could happen.
Kits were assembled just-in-time, shipped to whatever address the patient gave, timed to the draw window and the specimen's stability limits, and routed back to the lab intact. Normalized resulting across lab providers to give connected EHR experiences to digital health companies.
Butterfly Labs, 2020–2022. 300+ contracted phlebotomists, nationwide coverage, 2,000+ draws a month. Acquired.
Out-of-network claims arrive as paper. We ran OCR over the PDFs, built connectors to the payors, submitted the claims electronically, and automated the follow-up — because the submission is not the hard part. Getting a payor to actually respond is.
Get Franklin, 2019–2020. Bootstrapped, no outside capital, $10M+ in claims through the pipeline. I shut it down after concluding the model wouldn't reach venture scale, and started studyOS instead.
Product manager at Microsoft, 2016–2019. Azure SQL Data Warehouse, then the launch of Azure Synapse Analytics — which became the foundation of Microsoft Fabric after I left.
Right now I'm building personal tools for AI development.
Email is in my profile and DMs are open. hirokibutterfield.com is where I have recently decided I would write.
I'm a product person, so most of what I've shipped isn't code with my name on it and regulated software doesn't get open-sourced. Trying to build that muscle now.