About

Hi, I'm Sten Ilmjärv.

I wanted to know how people and diseases work, and I liked building things that helped answer that. Now I explore how AI works, what its future will be and how to stay human about it all.

Sten Ilmjärv

I started in gene technology and moved into bioinformatics early, because that was where biology met computers. My PhD in neuroscience looked at the brain through the same lens. A postdoc at the University of Geneva followed, as an embedded bioinformatician. All along I knew I wanted to build things, not only analyse data.

Then I joined Antion Biosciences, a gene and cell therapy start-up, as founding engineer and head of data science. We grew from 2 people to 17. The goal was a cure for HIV. Our platform could fine-tune how strongly a gene is expressed, which turned out to be useful well beyond HIV, so we took on cancer as well.

I developed a full-stack platform for our target sequence design, learned to use Kubernetes and built a custom orchestrator for it in Python. That was all pre-AI, or just at the time of the GPT-3.5 launch. I can't imagine how much quicker I would be now to develop the same thing, or what I would do differently. Times have certainly changed.

How I got here

Each stage built on the last.
None of them stopped.

Hover over the field, a skill or a stage.Tap the field, a skill or a stage.

How to read this

Each square is one day since 1 September 2003, 8,481 so far. Days run down a column, six weeks to a column. The colour is a skill.

The skills I began with are mixed evenly from the first day. A skill I picked up later starts as a narrow band in its own row. From there it spreads the way a drop of ink spreads in water. That is diffusion, and its width grows with the square root of time, which Einstein worked out for Brownian motion in 1905. So an old skill is everywhere and a new one is still close to where it started.

How much of each skill a period held is my own estimate, and every column holds exactly those shares. Which colour a single day gets is then decided by a fixed pattern built on the plastic number, a relative of the golden ratio. It spreads the colours as evenly as possible, so any patch of the field shows the shares the model gives. The dates are real. The mix inside a day is a model, not a timesheet.

  1. Gene technology (bioinformatics), 2003. Where it started.
  2. PhD, neuroscience, 2009. The brain, through bioinformatics.
  3. Postdoc, 2015. Embedded bioinformatician at the University of Geneva.
  4. Start-up, 2019. Founding engineer at Antion Biosciences, from 2 people to 17.
  5. Now, 2025. Building with AI and writing down what works.

Biology 2003 to 2017 · Bioinformatics since 2003 · Data science since 2009 · Software engineering since 2003 · Infrastructure since 2015 · AI since 2016

Now

AI moves too fast to follow from the outside, so I work with it directly. I build things with it, test what it can and can't do, write down what works and what doesn't, and take topics apart until I understand them. This blog is where that ends up.

I write it mainly for myself. If it helps someone else, or starts a conversation that goes somewhere I can't predict, even better.

What I write about

Anything that holds my attention long enough to try it myself. There is no plan and no fixed list. You can see where it has gone so far in the posts.

Get in touch

You can reach me on LinkedIn.