How I built language learning prototype with AI in one week

Some projects start with plans, others with pain points. This one started with a quiet thought in the middle of the night, the kind that doesn’t arrive with urgency but lingers long enough to make you open your laptop just to see what might happen.

I wasn’t trying to build anything serious or chase an idea that could turn into a product. It was more of a whisper than a plan, a simple curiosity to see if something small and personal could come alive, a thought that maybe, if I followed it, it could make learning feel lighter, or at least a little more human.

Problem

For the past months, I had been studying German some time to time as someone relocated to Germany a while ago but I was never able to prioritize it truly, but like anyone who starts a new language, I collected every kind of material that could possibly help me: books, magazines, articles which I would actually enjoy to read. I had folders full of them, both digital and printed, but somehow, I rarely opened any.

Every time I tried, I hit the same wall — new words I don't know everywhere.

Reading as a beginner is like hiking uphill with no end in sight: every few steps, you stop to catch your breath. That friction kills motivation before you even start.

And that night, as I looked at one of those very simple children books I bought, I caught myself thinking:

I wish there was an easy way to create notes and vocabulary directly from this text.

That one sentence wouldn’t leave my head.

And then, as it happens sometimes, the realization hit me so clearly.

AI can do that.

Solution

Before AI, my manual process for learning was slow but familiar:

I would open a book, magazine, translate every word I didn’t know, add each one to my notes, and later review them through the week, sometimes turning them into flashcards.

It worked, but it was exhausting.

It turned what should’ve been a joyful experience of discovery into a repetitive task.

That night, as I stared at another paragraph filled with unknown words, I realized I could teach AI to do exactly what I was doing, but faster, and smarter.

The idea was simple:

AI would read any given text; from a photo, a copied passage, or even something I ask it to generate which also came to my mind later and study it like I would.

It would analyze grammar structures, pull out new vocabulary, highlight interesting notes or fun facts, and then return all of that in a clean, structured way that I could store and reuse.

Something like this:

{
  "original_text": "In einem weit entfernten Universum lebte ein kleiner Junge namens Tim. Tim hatte schon immer davon geträumt, ins Weltall zu reisen und die Sterne zu erkunden. Eines Tages hatte er die Gelegenheit, an einer spannenden Weltraummission teilzunehmen.",
  "translated_text": "In a universe far, far away, lived a little boy named Tim. Tim had always dreamed of traveling into space and exploring the stars. One day, he had the opportunity to participate in an exciting space mission.",
  "source_language": "de",
  "target_language": "en",
  "notes": [
    {
      "note": "In German, the perfect tense is used to talk about completed actions in the past. It is formed with the auxiliary verb 'haben' or 'sein' and the past participle of the main verb. Regular verbs form the past participle by adding '-t', '-et', or '-en' to the verb stem.",
      "type": "grammar"
    },
    {
      "note": "'Ein Raumschiff' (a spaceship) is a neuter noun in German.",
      "type": "vocabulary"
    }
  ],
  "flashcards": [
    {
      "original": "das Universum",
      "translation": "the universe",
      "example": "Das Universum ist unendlich."
    }
  ]
}

The goal was to make sure the AI will always return a structured response that I could directly use, like a small, automated version of my study notes.

This would later evolve into an idea for storing them as flashcard objects, reviewing them with text-to-speech support, and turning each collection into a “language journal” of its own.

Of course, I knew it would require a lot of prompt tweaking to get the structure consistent, but as a starting point, it was already a huge leap forward.

I just didn’t know yet that I was about to face one big blocker that I hadn’t anticipated in the beginning but that came later.

The impulse to build

It wasn’t a planned idea but I had my notebook with me and I love making quick sketches to understand the idea. It took a few minutes and then I simply opened my laptop, and one line of thought led to another.

I used Next.js as our main web boilerplate library, connected it to OpenAI API, skipped authentication entirely, and picked the cheapest model available gpt-3.5-turbo just to see if it could work.

And within hours, it did.

By the next morning, I had a small working prototype: text input on one side, AI-generated notes and flashcards for vocabulary on the other. It was simple, raw, but surprisingly functional.

I even did not think about name before start building, it came afterwards: Langnotes. Short, descriptive, and easy to remember.

I bought the domain langnotes.app and shipped it online within 24 hours.

It wasn’t meant to be polished, but it already solved the very problem that inspired it.

That’s what I love the most about moments like this, when something you build moves faster than you do, and creation feels almost effortless.

You can try it yourself here ->

Polishing it

Over the next few days, I couldn’t stop thinking about it.

There was something pure about its simplicity for me; it did one thing, and it did it well and for quite some time I was meaning to experiment with AI more.

But as I used it more, new layers started to appear.

I began refining its identity:

• Creating and adjusting the logo (with ChatGPT) and color palette to make it feel more grounded

• Smoothing out UI interactions so it felt calm and encouraging to use
• Adding localization support for English, German, French, and Turkish

Localization felt especially meaningful. Language learning is deeply personal and cultural, and with AI, it’s possible to build something that reaches people in their native rhythm and with endless possibility.

Building in layers

I decided to keep everything on client-side storage for now.

There’s something beautiful about working this way; it’s fast, light, and gives you room to think without worrying about infrastructure too soon.

The next steps will come naturally:

• Add authentication to store notes safely
• Prepare for the mobile app version, which I believe will carry the real everyday value
• Keep the desktop app as an SEO base, and later bring that same approach to ASO when the app is live

Before expanding, I want to make sure the foundation feels right, clean, intuitive, and strong enough to grow.

One blocker I mentioned earlier, for example, was the copyright issue with translated text. When I tried using the first page of Harry Potter, the AI could easily generate study materials from it but refused to provide a direct translation. I’ll need to find a workaround for that, maybe by using a different model for translations or skipping full translations for now, since the core value lies in the study materials anyway.

It’s tempting to jump ahead, but after shipping the MVP, building something sustainable in long-term means learning to pause, to make space for the details that quietly shape the experience.

Reflection

Langnotes reminded me of something I often forget:

Building can still be pure when you follow curiosity instead of metrics.

It started as a small frustration on an ordinary evening and turned into a product I actually use in my own learning routine.

No deadlines, no external pressure, just a flow state that carried me through hours without noticing time pass.

A few days later, I even used Langnotes during a cozy autumn hike 🍂, pointing my camera at a German hiking sign, testing the photo-to-notes feature I had casually added.

It worked. Still requires some fine-tuning in the AI response but it can be done anytime.

And that simple moment, away from the desk, surrounded by trees, seeing the app I built do something real felt like the most honest kind of validation.

It’s not a big product launch or a success story. It’s a small reminder that the act of creation itself is where the joy lives.

The idea of Langnotes came from a personal need, and building it taught me once again that:

Our own problems are often the best ideas to build around.

And maybe, that’s the only kind of building that truly matters, the kind that starts from curiosity, grows through care, and ends up teaching you something about yourself along the way.