Ryohei Fujimaki became the youngest research fellow in NEC's history, then walked away to build dotData. His bet: automate the slow, unglamorous middle of machine learning that eats most of a data scientist's week.
Most people who reach the top of a 119-year-old institution do not leave it. Ryohei Fujimaki did. In 2015 he was named a research fellow at NEC - the youngest in the company's history, a title held by only six people among more than a thousand researchers. Three years later he handed it back to start over in Silicon Valley, chasing a problem almost nobody outside the field found interesting.
That problem is feature engineering: the slow, hands-on work of turning raw business data into the numeric signals a machine learning model can actually learn from. It is where data scientists spend the bulk of their time, and it is rarely the part anyone brags about. Fujimaki looked at it and saw something worth automating.
The company he built around that idea is dotData, headquartered in San Mateo, California. Its pitch is deceptively plain - let the machine build the features - but the technical claim underneath it is what drew attention from Fortune 500 buyers and research firms like Forrester before the broader market had the vocabulary for it.
Fujimaki earned his Ph.D. from the University of Tokyo in machine learning and artificial intelligence, and joined NEC Corporation in 2006. For the first stretch of his career he was a researcher in the classic sense - papers, algorithms, and the grind of moving an idea from a proof to something a business could run on.
The turning point came in 2011, when he transferred to NEC Laboratories America. There he led work on the semi-automation of machine learning and on heterogeneous mixture learning, and worked directly with NEC's global clients to deliver analytical systems that ended up in production across industries. The assignment that would define him was blunt: automate data science. He took it literally.
The prototype for dotData was not born in a startup. It was born inside a corporate research lab.
By 2015 the recognition arrived. The research fellow title at NEC is not a participation trophy - it went to six people in a research organization of more than a thousand, and Fujimaki was the youngest ever to receive it. Most careers would treat that as a summit. He treated it as a runway.
Founders usually start with a blank page. Fujimaki started with a working prototype and a corporate parent. Rather than quit and rebuild, he carved dotData out of NEC in 2018 - an unusual maneuver for a researcher at a company that predates the transistor. The carve-out let him keep the technical core while giving the business the freedom of a Silicon Valley startup.
The wager was specific. If you can automate feature engineering, you compress the part of a machine learning project that swallows the most time and the most senior talent. Do that well and you widen the pool of people who can build a useful model from a handful to a whole department.
The vision of dotData Enterprise has always been to make predictive analytics more approachable to a wider spectrum of users.
Where the time goesThe reason feature engineering is worth a company is arithmetic. Across a typical machine learning project, the raw modeling is a sliver. The bulk of the effort sits in preparing, wrangling, and shaping data into features - and then in deploying whatever comes out. The chart below is illustrative of how that time tends to split, and it explains where Fujimaki aimed dotData's automation.
This is also why Fujimaki has spent years pushing back on the loudest fear in his field - that automation will make data scientists obsolete. His answer is consistent and unromantic.
AutoML 2.0 platforms and data science automation are not going to eliminate the need for data scientists.
The point, as he frames it, is productivity rather than replacement. "With automation platforms, data scientists can become more productive by accelerating the development of AI models and automating the development of feature tables," he has said. The machine takes the repetitive load; the human keeps the judgment.
The founder's operating adviceFujimaki's caution about AI adoption is the kind that comes from having shipped systems, not just theorized about them. He is skeptical of the impulse to experiment for its own sake.
"The most important part of embracing automation is to understand where it is likely to provide the greatest benefit and what the greatest risks of failure are," he has said. And more pointedly: "The biggest risk of failure comes when companies try to experiment with AI and begin without clear, compelling, measurable use cases." It is a founder telling his own market to be disciplined - unusual advice from someone selling the tool.
The current chapterThe idea that made dotData has kept evolving. In 2024 the company shipped Feature Factory with generative-AI-assisted feature discovery and added a GenAI "Use Case Advisor" to help teams pick problems worth solving. In 2025 it followed with Feature Factory 1.3 - AI-powered column enrichment, automated text processing, expanded large language model support - and dotData Enterprise 4.0, a full redesign built on the Feature Factory engine.
There is a quiet vindication in the sequence. Fujimaki was automating the middle of the machine learning pipeline years before generative AI turned "automate the workflow" into a universal slogan. When the wave arrived, his company already had the plumbing.
Away from the product, he writes as a Forbes contributor and has authored a steady stream of articles on data science automation - the same measured, use-case-first voice that shows up in his interviews. The through-line from the Tokyo lab to the San Mateo office is that he keeps building for the practitioner, not the hype cycle.
What Fujimaki is chasing has not really changed since that assignment at NEC Labs. Take the slow, expensive, expert-only work of turning data into decisions, and hand as much of it as possible to a machine - so that more people, in more companies, can do it. He gave up a title almost no one earns to go do it in the open market. So far the market has agreed it was worth the trade.