CoA Nexus, Inc.

The premise is that a researcher already publishes their resume, constantly, in public. It is called their body of work. Papers and patents are, if you squint, an unusually honest form of self-description: hard to fake, dated, peer-reviewed, and stuffed with the specific words that describe what a person can actually do. CoA Nexus built an AI that reads that trail - by its own account, more than 30 million research papers and over a million patents - and turns it into a map of who is good at what.

That map is the product. Its flagship platform, CoA Researcher, matches R&D talent with the companies that need it, and the company describes it as the largest researcher-matching platform in Japan. The reach spans the unglamorous, high-stakes fields where hiring is hardest: biotechnology, chemistry, medical science, semiconductors. A newer product, CoA Atlas, points the same engine at recruiters, letting them source from a pool the company sizes at roughly three million R&D professionals.

The clever bit is not that the software finds people. Lots of software finds people. It is that CoA Nexus is willing to say the quiet part out loud: keyword search on a CV is a bad way to judge a scientist, because a scientist's best keywords are the ones they never bothered to put on the CV. Reading the work is slower, harder, and more expensive. It is also, plausibly, correct.

There is a second, softer idea running underneath. CoA Researcher does not only fill permanent jobs; it also books researchers for side work and part-time engagements. This sounds small. It is not. A great deal of specialized talent is locked inside a single employer, doing one thing, while a startup three train stops away would pay handsomely for four hours of that exact brain. Making that trade legible - turning a scientist into someone who can moonlight the way a designer does - is the kind of unglamorous plumbing that quietly changes a labor market.

None of this requires you to believe the grander framing, which the company prints on its own homepage: "Accelerate Deep Tech Revolution, Cultivate a Better Future." That is a lot of capital letters. But strip the slogan and a reasonable business remains - a matching layer for a market that is large, opaque, and badly served by the tools everyone else uses.

Founder and chief executive Kota Nozaki arrives at this from a genuinely odd direction. Before starting the company in 2020 - then named Srust, Inc. - he led project finance and fund formation in renewable energy, the sort of work measured in hundreds of billions of yen and long dam-shaped timelines. Somewhere in there he apparently concluded that the harder bottleneck was not capital or kilowatts but people: the right scientists, findable and mobile. The rename to CoA Nexus in 2024 marked the shift from a single product to a small stack of them - Researcher, Atlas, plus commercialization support (CoA Accel) and a media arm of researcher interviews (CoA Study).

In October 2025 the thesis got a price tag. CoA Nexus raised roughly ¥300 million in a Series A, drawing in a syndicate that mixes the strategic and the financial: advertising conglomerate Hakuhodo, which also entered a capital and business alliance; Lotte Ventures Japan; Innovation Engine; and BlueSeed, alongside debt from financial institutions. That brings total reported funding to about ¥500 million. The money, the company says, goes toward expanding the researcher database internationally and building the features to match across borders - which is what the Menlo Park office is presumably for.

The risks are honest ones. Inferring skill from publications is a genuinely hard inference problem, and being confidently wrong about a scientist is worse than being silent. A seven-person team maintaining a 300,000-person index is leverage right up until it is a staffing constraint. And a two-continent strategy doubles the surface area of everything. But the core wager is a good one, and refreshingly concrete: the resume was always a lossy compression of a person's work. CoA Nexus just decided to read the original file.