Michael Castelein
Dutch-American · Data & AI
I'm Michael, a Dutch-American consultant running ML Ventures. Half the work is data: analytics, BI, and the pipelines underneath. The other half is training teams to use AI inside the work they already do. On the side I build language-learning software and small tools. I do all of it from four homes across three countries.
Right now: Netherlands, UTC+2
A globe showing the annual route between Chaozhou, Kunming, The Netherlands, Atlanta.
Drag to spin. The arcs are the real route.
What I do
One is building the data a business runs on. The other is getting the people in that business to use the tools already sitting in front of them.
01 / The engine
Analytics, BI, analytics engineering, data science and ML. This is what the agency has run on for years, with two subcontractors and no office in any of the four countries.
02 / The newer half
Most companies have already bought the AI tools. Almost nobody uses them well. I run hands-on workshops with teams to fix that, working in their own files and their own workflow. No slide deck about the future of AI.
Projects
Products, and tools that started as a problem I had. Some are live in the browser, some run on my own machine.
Practise a real IELTS exam, get a band score back in seconds.
An IELTS prep service covering all four sections: listening, reading, writing and speaking. You answer real questions from past exams, an AI examiner marks the answer and explains the band score, and a dashboard tracks where you are improving and where you are stuck. Next.js, with Claude doing the marking.
Drag coffees onto a day and watch what your blood does.
A sandbox for understanding caffeine. Drop drinks on a timeline and the curve, the bedtime number and every panel below redraw live. It runs a one-compartment pharmacokinetic model with first-order absorption, so a dose peaks about 45 minutes in and then halves on your own clearance rate. One HTML file, no server, works offline.
One Mandarin app instead of the seven I was juggling.
A Mandarin learning app built around a spaced-repetition engine. Everything else hangs off that same vocabulary and progress model: an AI reader that generates passages at your level, listening drills, mock HSK exams, character decomposition, and a tutor you can talk to. Next.js and Supabase, with an FSRS scheduler at the centre.
Now
Updated August 2026
Building
ML Ventures is my company. The consulting pays for the rest, and the rest is why the consulting stays interesting.
Data, analytics and AI workflows on LinkedIn. Travel on Instagram, YouTube Shorts and TikTok. A following on HelloTalk that grew out of language content I never planned to make.
The tools I wanted while learning Mandarin and could not find: SRS notecards, audio cards, adaptive readers. Built for my own studying first, which is why they are any good.
Lina runs a Shopify store selling clip-on nails, sourced from China. I help with the parts that are spreadsheets.
The long idea: places in Chaozhou and Kunming where people come and live in the language rather than study it. Years away, not months.
Where
The same route, in the same order, every year: Chaozhou, Kunming, the Netherlands, the United States. Chaozhou and Kunming are both in China and share almost nothing. Different food, climate and language, with a thousand kilometres between them.
01 / Family
January to March · China · 23.7°, 116.6°
The reason the loop exists at all.
Chaozhou is my wife Lina's hometown, a Teochew-speaking city in eastern Guangdong that almost no foreigner ends up in. The language there is not Mandarin and mostly is not written down. We spend the first three months of every year here.
02 / Deep work
April to July · China · 25.0°, 102.7°
Eternal spring, nineteen hundred metres up.
Kunming is called the spring city because the weather refuses to do anything interesting. High plateau, mild all year, thin clean air. This is the longest stint and the one where most of the building happens.
03 / Roots
August to October · Netherlands · 52.1°, 5.3°
The Dutch half, in the good months.
Half my passport and most of my twenties. I studied here and worked here: BI at HousingAnywhere in Rotterdam, then data engineering at a consultancy. I started the agency here, before it stopped needing an address.
04 / The other passport
November to December · United States · 33.8°, -84.4°
Closing out the year on the American side.
The other half of the hyphen, and where the year ends: holidays, family, and the annual reset. Most of what the US means on my CV happened earlier, an internship at Southwest and an actuarial co-op at John Hancock in Boston. These two months are the part that repeats.
Before this
Off the clock
I founded a table tennis association and chaired it. Still the sport I would keep if I could only keep one.
Former chairman of a badminton association, and still the second racket in the bag.
Working toward a first triathlon, with an Ironman somewhere further down the list.
I model things that do not need modelling and I enjoy it. This is a sincere entry.
Languages
Get in touch
If any of this overlaps with something you are doing, a data problem, getting AI to land in your team, learning Chinese, or you just liked the map, say hello.
Message me on LinkedInI answer from whichever timezone I am in. It changes four times a year.