The Question

"What if the machine can be what I can't?"

I'm a senior software developer from Finland. I've been writing code professionally for over 12 years, all self-taught. No university, no computer science degree. I learned to code because I wanted to build things, and the Finnish education system didn't have a path that matched how my brain works.

Because my brain works... differently.

I have ADD. The kind where some days the operating system boots and I deliver three weeks of work before dinner. Other days it doesn't boot at all, and I stare at the screen wondering if I ever knew how to code. There's no middle setting.

The upside of ADD is hyperfocus. When something grabs my attention, I don't just learn it — I disappear into it. The downside is that I can't choose what grabs me.

I know myself well enough to know I shouldn't trade manually. I'm impulsive, emotional, and inconsistent. So the question became: what if I build something that doesn't have my flaws? A system that's patient when I'm not. Consistent when I can't be. Awake when I'm asleep.

In 2024, Hidden Markov Models grabbed me.

The Search

I started with LSTM neural networks — the popular choice for time series prediction. Months of work. Didn't deliver. The answer turned out to be one of the oldest tools in the statistical toolbox — and one you can actually understand, not a black box.

HMMs are 120 years old. Andrei Markov invented the underlying math in 1906. They've been used to decode speech, sequence DNA, predict weather, and — rumor has it — make Jim Simons very, very rich.

The last point matters. Neural networks will always give you an answer. HMMs will shrug and say "this looks like noise." I trust the model that admits uncertainty.

Then the ADD kicked in. My brain runs on "what if" loops. Most of them go nowhere. But occasionally, one connects to another, and another, and suddenly there's a system.

That's the ADD pattern: try, fail, try, fail, fail, fail, "wait, what if I..." — breakthrough. Repeat until it works or you run out of energy.

The Journey

This system wasn't built in a weekend. It's taken almost two years to reach its current state — hundreds of hours, and more failed experiments than I can count.

Each tier obsoleted the previous one. Code was deleted, rewritten, deleted again. The current 5000-line trainer script is the survivor of many purges.

Finland is not Silicon Valley. We don't have a culture of "move fast and break things." We have a culture of "think carefully, build properly, don't brag about it."

Sisu — the Finnish concept of stubborn resilience — is probably the best description of how this system got built. Not through brilliance. Through refusal to stop.

Also, Finnish winters are long and dark. You need a hobby. Some people cross-country ski. I train Hidden Markov Models. Both are endurance sports.

It's never ready. Every week I find something to improve. The trainer runs 24/7, mining for better models while I sleep. When I wake up, there are new champions to evaluate. When I have an idea at 2 AM, I can test it by morning.

Ei se ole koskaan valmis. Se vain paranee.
(It's never finished. It just gets better.)

The Edge

I'm not a mathematician. I don't pretend to be. But I understand probability, and I understand one thing very clearly: if you have a small edge and you repeat it thousands of times, you win. Not every time. But always in the long run.

Casinos don't worry about individual hands. They worry about sample size. I built a casino where I'm the house.

The system doesn't predict the future. It detects statistical patterns and bets that they'll continue. When they don't — and they won't, sometimes — the losses are small. When they do, the wins are bigger.

Markets are asymmetric in the long run — they drift upward over time. Being on the right side of that drift, and stepping aside when it reverses, is enough.

The law of large numbers does the heavy lifting — over thousands of trades, a small edge becomes inevitable. But you have to survive long enough to get there. Markets go through regimes — bull, bear, sideways, crisis. No model wins them all. The goal isn't to be right every time. The goal is to survive the bad regimes and compound through the good ones.

"Smoothaus > Maksimointi — Selviytyminen > Voitto"
(Smoothing over maximizing — Survival over winning)

Live

The system has been trading live since February 2025 — just before Trump's tariff dip hit the markets. Real money. Real trades. Real emotions when the market opens and your models disagree with your gut.

The first year was profitable — roughly 20% gains, 54% win rate, 1.41:1 profit-to-loss ratio. Not every month is green, but so far it's been beating the S&P 500. One year of live results later, we shifted to a more defensive strategy.

Now running Titanium Age — the sixth major iteration.

Every model validated against 2007-2015 data it never saw during training. If it can survive the 2008 financial crisis without too big of a drawdown — and what counts as "too big" depends — it earns its place. Most don't.

I trust backtests that survive out-of-sample validation. But I trust live trading results more. One year of live data is worth more than ten years of backtesting.

Longer term, I'm considering opening the signal server in some form. The architecture already supports it — the system broadcasts signals via WebSocket, and connecting additional clients is technically trivial.

But first I need more live trading history. Months of it. Maybe a year. Showing backtests is easy. Showing a live track record that matches the backtests — that's the hard part, and the only part that actually matters.

The Honest Part

This might not work.

Past performance doesn't guarantee future results. Markets change. Black swans exist. Models degrade. I could be fooling myself with sophisticated backtesting.

But here's what I know: the math is sound, the validation is honest, and the system was built by someone who's failed enough times — in code and in real life — to know the difference between hope and evidence.

The edge is small. The sample size is large. And the machine never sleeps.

Katsotaan miten kay.
(Let's see how it goes.)