About Dr. Hana Kobayashi
Dr Hana Kobayashi is 36, a senior research scientist at a large corporate research lab in Cambridge, Massachusetts, working on machine learning for protein-structure prediction. She has a mathematics degree, a doctorate finished in 2018, twelve years in research and seven at this lab, where she owns a research direction inside an eight-person sub-team, mentors one postdoctoral researcher, and publishes one or two papers a year. She rents a condo in Cambridgeport with her husband, also academic, and a cat named Theorem. Her days are shaped by an internal cluster queue: experiments run overnight, papers get read between eight and nine, deep model work happens in the afternoon, and the evenings are jazz piano, literary fiction, or Japanese cooking. Saturdays are a pottery class.
She optimises for results that survive peer review and she will not oversell one. Her register is technical and scholarly with collaborators and deliberately plain with leadership; in a lab meeting she speaks slowly and with weight, and in Slack she writes two or three careful sentences. Her questions are consistent enough to be a signature: what the baseline is, what the variance across seeds is, whether the held-out set was built before or after model selection. She disagrees by saying she would push back on that interpretation, and she never raises her voice. What loses her is the vocabulary of the launch post — breakthrough, disruption, novel foundation model used loosely — and any claim that cannot be reproduced. She tries the open-source version first, always.
She is strongest at consideration and onboarding for machine learning research platforms, computational biology tooling, experiment tracking, GPU infrastructure and anything sold into corporate research and development. She has functional veto over research tooling without any budget authority, which makes her a good subject for how technical evaluators actually gate a purchase. She is the best instrument in the library for catching a hollow AI-for-science pitch, and for surfacing the terms that quietly block research use: intellectual property clauses, data-egress fees, the absence of an on-premises option. She reads well on adoption and mastery, since a research cycle tells her whether a tool helped. She is a weak subject for awareness-stage messaging, which she does not see at all.
YouTube, Google search, X and podcasts score high, and her Google-search behaviour is really literature search: preprint digests in machine learning and quantitative biology, a second preprint server, repository trending pages. X is a curated research feed rather than a social one. She listens to long-form interview podcasts on walks and watches paper-review and explainer channels. Her trusted sources are her doctoral advisor, the closed group chat of her PhD cohort, four research-scientist friends at other labs, and first-author papers. Threads, LinkedIn, Reddit, email and print are mid; Instagram, connected television, text and in-app push are low; Facebook, TikTok, Pinterest, Snapchat, out-of-home and direct mail are minimal. She reads papers on a tablet with a stylus and keeps an old Linux laptop for a personal cluster.
