I play traditional Irish music. Not professionally — it’s a passion, not a career. But trad has a particular challenge that every learner eventually hits: the sheer volume of tunes. Hundreds of jigs, reels, hornpipes, polkas. Learning them by ear. Remembering them consistently. Knowing when you’ve actually got one into your fingers versus when you think you have.
I wanted a simple tool to help with this. A trainer. Something that would quiz me on tunes I was working on, let me track what I knew, and keep me honest about the gaps. I looked around and nothing quite fitted. So I built one.
This is the story of how that happened, what I used, and what it unexpectedly taught me about where AI development is right now.
What the Trad Tune Trainer Actually Does
The app is simple by design. You can add tunes you’re learning, tag them by type (reel, jig, slip jig, and so on), mark your progress, and use the trainer mode to quiz yourself on what you know. It’s not trying to teach you how to play — that’s what sessions, YouTube, and your teacher are for. It’s a memory and progress tracker built around the way trad musicians actually learn tunes: repetition, recognition, and gradual ownership.
You can try it at trad-tune-trainer.lovable.app. It’s free and works in any browser.
How It Was Built: Lovable and the AI-Assisted Development Stack
I built it using Lovable, an AI-powered web app builder. If you haven’t come across it: you describe what you want in plain language, and Lovable’s AI generates working React code that you can edit, refine, and deploy. No wrestling with a blank file. No context-switching between design, code, and infrastructure. You talk to it like a collaborator, iterate on what it produces, and end up with something real.
This is sometimes called “vibe coding” — a term that sounds flippant but describes something genuinely new. The idea that someone with domain expertise but limited development time can describe a problem and get a working solution in hours rather than weeks. It’s not magic, and it’s not always smooth. But it fundamentally changes the equation for what’s practical to build.
What I Actually Learned
AI is a collaborator, not an autocomplete. The best results came when I had a clear picture of what I wanted the app to do and could describe it precisely. Vague prompts produced vague results. Specific prompts — “I want a card that shows the tune name, type, and a confidence rating I can update with a single tap” — produced something close to what I needed immediately.
The bottleneck shifts from writing code to making decisions — and those are questions the person closest to the problem is best placed to answer.
The bottleneck shifts from writing code to making decisions. Traditional development has you spending a lot of time on implementation. With AI-assisted tools, implementation becomes fast. The time goes on clarity: what does this feature actually need to do? How should data be structured? What does the user experience need to feel like? These are product questions, not technical questions, and they’re questions that the person closest to the problem is best placed to answer.
Iteration speed changes what you’re willing to try. When a feature takes two days to build, you think hard before committing to it. When it takes twenty minutes, you try it and see. This changes the whole creative dynamic. I added features I’d never have prioritised in a slower process, discarded things that seemed good in theory but felt wrong in practice, and arrived at a better product faster because the cost of being wrong was so low.
Domain knowledge still matters enormously. Lovable can generate an app. It cannot know that trad musicians learn tunes by ear and care about sets (groups of tunes played together), not just individual pieces. It cannot know that “confidence rating” means something specific in this context — not how much you like a tune, but how reliably you can recall it under session conditions. That knowledge came from being a player. Without it, the app would have been generic and less useful.
Domain knowledge still matters enormously. The AI can generate the app. It cannot know what the app needs to feel like to the person who will actually use it.
Why This Matters Beyond a Personal Project
This is a passion project. But the technology and the process behind it are the same ones I talk to clients about every week.
The businesses I work with often have a clear problem they’d love a tool for — a custom reporting dashboard, an internal knowledge base, a simple client-facing tool — but assume building it is out of reach. Too expensive. Too slow. Requires a development team they don’t have.
That assumption is increasingly wrong. Generative AI hasn’t just made existing software smarter. It’s compressed the gap between “I have an idea” and “this works” in a way that’s real and practical right now.
The trad tune trainer is a small example. But the lesson generalises: the constraint on what you can build is no longer primarily technical. It’s clarity of thought. Know your problem precisely, describe what you need specifically, and the tools available today can take you further than you probably expect.
Try It, and Tell Me What You Think
The app is live at trad-tune-trainer.lovable.app. If you play trad — or know someone who does — have a go. I’m still adding to it, and feedback from actual musicians is the only thing that makes a tool like this better.
And if you’re a business owner thinking about something you’d love to build but have never quite started — let’s talk. It might be closer than you think.