HANGAR · RAW MATERIAL: PASSION · DWG SD-001
That is how a method works — not magic.
Case study: SoundDNA — how a personal need became an international research platform measuring how people really listen.
PASSAGE PROTOCOL · KARBONADO METHOD · ALL STAGES CLEARED
01RAW MATERIAL ON THE TABLEAn amplifier that missed my ear, the SoundsBetter studio, hundreds of conversations with audiophiles✓ 02WILL IT FLY?Hypothesis: individual profiles of sound perception exist and are measurable — confirmed in 2026 on my own hearing✓ 03STRUCTURAL BLUEPRINTA research methodology with scientific rigor — before a single line of code✓ 04BUILD IN THE WORKSHOPA platform in three languages; automated tests guarding the product's promises, including the privacy promise✓ 05ENGINE AND TICKET MACHINESoundDNA Playlist Analyzer, SoundDNA Compatibility, SoundDNA Casting, SoundDNA Academy — reasons people board on their own and bring the next ones✓ 06FLIGHT AND HANGARMonitoring, backups, a load test before launch — and the telemetry you can see below✓In this hangar I only show machines I can vouch for with my signature or my stake — two of my own, and one, brought in, where I am the architect of the whole and a co-owner.
It all started with a search for an amplifier that would match my hearing. Every one I auditioned had gaps in its sound — not objective flaws, but mismatches with what my ear was looking for in music. Since the market had no answer, I built one: I started re-tuning amplifiers until they played in a way that pulled me into listening for hours.
I did it well enough that audiophile friends began bringing me their gear. That is how SoundsBetter was born — a small studio tuning audio electronics. And it was there, tuning unit after unit, that I saw something more important than any circuit.
The end result always earned a “wow”. But the road there exposed the same barrier every time: the client and I were describing the same sound in languages that refused to meet. Audiophile vocabulary is, in theory, richly developed and full of beautiful, pictorial concepts — but each person describing the same picture may hold a different one in their head. Tell someone “it looks like a cat” and everyone will see a cat in their imagination — but will everyone's cat look the same? That is exactly how the language of music perception works: the descriptions are beautiful and vivid, yet they do not operate on what happens inside the listener's head — they assume that all the cats in the world are identical and everyone sees the same cat.
I realised that even doing technically flawless work, I could still miss the person's expectations — because I did not know HOW they heard. Not what they listened to, not on what. How.
This was a gap at the level of a fundamental need: no tool existed that described individual profiles of sound perception. I hypothesised that such profiles exist and can be measured — and I decided to test it rigorously, not with an internet quiz.
Stop guessing why people do things — understand how they think. This is the sentence my whole practice stands on. In SoundDNA I took it literally and built an instrument that measures it.
The understanding that everyone hears differently had been maturing in me for years. But between “I know” and “I have lived it” runs a border you cross only in one direction — and I was taken across it by an event I would not wish on anyone.
At the beginning of 2026 I suffered a sudden hearing loss — the tests showed a deficit of around 90%. To this day nobody knows what caused it, and in this story that does not matter. What matters is what happened next: my hearing came back. But it came back different.
The same amplifier. The same speakers. The same tracks, listened to hundreds of times before. And my hearing memory kept reporting a mismatch: things appeared that I had never paid attention to — or simply had not heard; things I knew by heart disappeared. I began discovering my entire music collection anew.
For three years I had understood THAT people hear differently. Now I had HEARD differently myself — in an experiment nobody volunteers for, yet with the variable isolated perfectly: neither the gear nor the music changed. The listener did.
That was the ignition. A hypothesis that had matured for three years received its proof written on my own sense — and that is why three years of maturing turned into six weeks of building. In my workshop I call such experiences a scraped knee: it hurts, and it leaves knowledge that cannot be bought (more on that approach on the About page). SoundDNA today measures exactly what I lived through: a difference that resides not in the equipment, but in the listener.
The hypothesis became a methodology, and the methodology became SoundDNA: an independent, non-profit research platform that measures a person's sound-perception profile in a ~15-minute test and builds open knowledge about how the world listens. Today it runs in three languages (Polish, English, German — produced through an adaptation process with independent translators, not a “quick translation”: the point was that in every language everyone describes the same cat), with a set of tools around the result: listening-compatibility comparison between two people, streaming-history analysis set against the profile, and public statistics with privacy-protecting thresholds.
Why non-profit? Because it is a structural decision, not a sentimental one. The credibility of perception research requires that no one can ask: “and who are you selling this data to?”. The answer is built into the architecture: the test collects no personal data, research data is separated from it by construction, and the absence of a profit motive on data is the foundation of trust — the fuel this machine flies on.
A test that only measures would have been interesting on its own — and easy to forget. SoundDNA is built as a platform on purpose: four tools sit around the result, each one designed to hand people to the next. That is a different kind of thinking from “we built the test, it works, it collects data, that's it” — it keeps asking what else would give people more value back, and pull more people in. Not a lone feature. An engine of tools feeding each other — all bolted into one block: the Test itself.
FOUR TOOLS · ONE ENGINE · STAGE 05 IN PRACTICE
Turns a raw streaming export — the file the platforms hand over but no ordinary program can open — into a shareable portrait: the artists, genres and hours behind how someone really listens. Then it lays that against their test profile, because how a person listens and what they listen to are two different stories, and the gap between them is the interesting part.
A portrait people post is a portrait that brings the next test-taker — the advertising the platform never has to buy.
Two, three or four profiles side by side, mapping where listeners' ears meet and where they split — not who is right, but which layer of the same track each one hears. Its trick is that it is useless alone: to get anything out of it, a user has to pull in the next person.
One test-taker recruits a friend; the friend brings ten. The growth is not bolted on — it is the mechanism.
Builds a listening panel the way a director casts — by measured profile, not by who had time. An organiser pastes up to a hundred results, sets the seats, and picks the most diverse panel, the most aligned, or a full set of the four listening types; the tool chooses and shows its reasoning. And the list they paste never reaches us — the difference between “we promise not to sell” and “we have nothing to sell.”
Proving the profile has value beyond one person's curiosity — so it pulls in organisers, clubs and editors, each arriving with an audience of their own.
The other three work on the result; this one works on the person. Two hundred calibrator tracks across the four listening types and three levels, each training one skill of attention — free, no account, endless. It turns a one-time test-taker into a returning student, and sends them back to the test to see how their ear moved.
The cylinder that keeps firing after the test ends — the retention loop the other three could not close.
Each tool is a cylinder in this engine, and every cylinder added is not a spare — it is more power and more reliability at once. Four cylinders do not just keep the crankshaft turning; they turn it faster, harder, with torque to spare if one misfires. Starve one of fuel and that cylinder stops — the engine still runs, only rougher, only slower. That is the extra-nines math I used to price into major IT builds: every nine you add tacks another zero onto the invoice — and here, every cylinder you add tacks another zero onto the user count. That is what stage 05 means in practice: not one feature, but a machine built so every added part multiplies what the whole engine can do.
On the delivery side, the project combines existing tools with programming the ones that did not exist — including a research instrument built from scratch, a comparison engine and a statistics publishing system.
What may interest you most is not the technology but the process — because this is precisely the standard I build to in stage 04.
I delivered the project in tandem with two specialised AI agents: an architect-diagnostician (design, decisions, quality control) and an executor (implementation) — with a clear division of roles, working rituals, documented decisions and hard safety rules. I acted as product manager and the final arbiter of every decision.
One thing I want to say plainly: AI did not lower the rigor here — it accelerated the work while keeping full control. Automated tests guarded the product's promises (including a privacy promise encoded in tests), and hands-on human testing preceded the acceptance of every stage. Plus backups, monitoring and a load test before launch.
The result: a scope of work that a classic team would deliver in two to four quarters — depending on how well the team members fit together and how communication with the client runs (I know, because I managed a software house) — closed in six weeks.
FLIGHT TELEMETRY · READ FROM THE PLATFORM · AGGREGATED DATA ONLY
These numbers are not a screenshot — they are a cockpit read-out. The platform publishes aggregated statistics only, with privacy-protecting thresholds: this is what the anonymity architecture from stage 04 looks like in practice.
In full honesty: those six weeks were work in deep flow — focused exclusively on this one project, sleeping 4–6 hours a night, with every other activity switched off for the duration. That pace has its price, and I won't pretend it comes free.
In the Method, pace is chosen like a bungee rope — for the weight and the goal. Not every project needs a sprint; this one did, because I was jumping on my own rope.
And one more number — more important than all the others: from the idea to its execution, three years passed in my case. That was not wasted time — I was building the knowledge of how to do it well and preparing the tools — but I know that this is exactly what everyone with a good idea struggles with: not a lack of vision, but the road from vision to the first working version.
And here is the difference between your road and mine: I already have those three years behind me. The tools, the method and the working rituals are built and battle-tested — on a project of international scale, not on slides. For your idea, that stage does not have to take years.
If you have diagnosed a real need — even a niche one, even one that feels “only yours” — the distance from an idea to a working, multilingual product is shorter today than intuition suggests. The requirement is not a big team but the right method: a sharp diagnosis of the need, hypotheses instead of assumptions, rigor where it builds trust, and well-orchestrated collaboration between people and AI.
And once your raw material passes the trial, the question stops being “will you get your project built”. It becomes: how much time will you save through this collaboration — and how high my knowledge and my tools will carry your project into the sky.
The first conversation costs nothing and promises one thing: I will tell you honestly whether I will build a machine that flies.