From Prototype to Platform: Building Tchaikovsky
Tchaikovsky didn’t start as a company. It started as an idea on my laptop, in a small room in Chiang Mai, Thailand. At the time, it was just me experimenting — trying to see if I could connect a language model to ABC notation and have it generate something musical. Not just noise — actual compositions.
I had no big roadmap, no investors, no team — just a rough prototype and a question:
Can AI actually write music in a way that feels editable, real, and useful?
The First Build
The first version was scrappy. A basic Python script that took a prompt, passed it to DeepSeek, and converted the output to ABC notation. From there, I used music21 to translate that into MIDI, which I’d manually load into GarageBand just to hear if it worked.
It did.
Sort of.
The generations weren’t perfect, but they had structure. I could see the possibility — especially when I tweaked the prompts. That’s when it clicked: this wasn’t just AI music. This was AI composition. Symbolic, editable, and most importantly — in my control.
So I started building something real.
A Browser-Based DAW — By Myself, at First
I designed and built the early DAW interface in React. The backend came together in Flask, then FastAPI. The front end started supporting MIDI uploads, piano roll editing, and export features. I spent weeks buried in bugs related to sound playback, MIDI timing, file conversion, and UI quirks. There were many days it felt like nothing worked.
But little by little, it did.
I got the app working on desktop, then wrapped it for iOS and Android. I wired in Firebase for user accounts, set up project saving, connected cloud storage — and kept refining the generation flow.
Eventually, it stopped feeling like a prototype. It started to feel like a platform.
Collaborators, Not Just Code
Amber came on board early — not just as a partner in business, but someone who understood what it meant to build something in Thailand, for real. With her background and support (and a lot of help from Jake at Green Light Studio too), we figured out the logistics that come with starting a company here: the 51% ownership requirement, the startup visa maze, the hosting plans, the government forms.
It wasn’t glamorous, but it was real. We weren’t pitching VCs. We were wiring up servers at midnight.
Building the Right Kind of AI
What makes Tchaikovsky different is the decision we made early on to treat music like language. Not just notes, not just patterns — real structure. The AI doesn’t spit out WAVs. It writes ABC notation, which we convert and display directly in the DAW.
That means the music is editable. It’s transparent. You can see what the AI made, tweak it, extend it, or rebuild it.
This gives creators actual control — something most AI music tools still don’t offer.
A Real Product. Now What?
As of 2025, Tchaikovsky is live — fully functional in browser, on iOS, and Android.
You can generate tracks from prompts, edit them in the piano roll, save and export them as MIDI, WAV, MP3, or sheet music.
But we’re still building.
Next comes real-time collaboration. A full WAV editor layer. A better AI system that understands dynamics, phrasing, and emotion. And yes — eventually, we’ll raise funding to support that scale. But for now, we’re staying small, focused, and close to the users.
Final Thoughts
This started as a solo experiment. Now it’s a product people use — and a platform we’re proud to keep improving.
If you’re reading this, whether you’re a musician, developer, educator, or just curious — thanks for being here. We built Tchaikovsky to give more people the tools to create, explore, and compose with AI.
There’s a lot more coming. But this is how it began.