Personal project / Iteri / Learning & AI
What began as a small way to study Japanese became my ongoing experiment in building software with AI — and in discovering how much work still happens between an idea and something genuinely useful.
01 / The idea
Before Iteri, I studied with a little collection of apps: Anki for flashcards, dictionaries for unfamiliar words, translation tools for sentences, and ChatGPT whenever I couldn't quite understand an explanation. Each was useful. Moving between them wasn't.
My first idea was modest: take the exercises from textbooks I already owned, answer them on screen, and let AI grade my answers and explain why.
But once that worked, I started wishing for more of the book in the same place — reading, vocabulary, grammar and listening. Iteri grew from that habit of noticing one more thing I'd like to change.
02 / The possibility
For a long time, AI was something I asked questions or used for a small piece of an automation. As the technology advanced, I began to see a different possibility: I could try building a whole application, even without a software-engineering background.
What surprised me was how much my experience designing business processes helped. I couldn't write the code myself, but I knew how to think through inputs, decisions, handoffs, exceptions and what should happen when something failed.
So I didn't want to simply ask AI to build a feature and accept whatever came back. I needed a way to guide the work and check the result.
The working rhythm I follow
Not a straight line. A failed test often sends me back to rethink the plan.
03 / The reality
AI can produce something that looks promising surprisingly quickly. Making it work the way I imagined is much harder. Some frustrations are visible; others only appear when I try to study with what I've built.
The visible frustrations
I've asked for fewer boxes in the UI and somehow ended up with more. A feature can technically work but feel painfully slow. Fixing one bug can introduce another. Sometimes the first version is just the beginning of a much longer conversation.
A working feature isn't necessarily a good experience.
The less visible problem
Textbook questions, answer keys and listening transcripts often live in different places. AI could connect the wrong sources, then confidently explain an answer that belonged to another question. That was harder to spot — and more serious for learning.
A convincing answer isn't always the correct one.
At first, I hoped AI could figure out all those connections on its own. I changed my approach: treat questions, answers and transcripts as separate sources, add checks and confirmation steps, and keep answers hidden until a learner submits a quiz. Some of this is still evolving, but it changed how I think about reliability.
04 / Learning by using it
Iteri is still an early product. But I already use it to study Japanese, especially reading and vocabulary, while I continue working on quizzes, listening and the links between book materials.
Being my own learner means I notice problems immediately: an explanation that doesn't help, a missing passage, a step that takes too long. I close the lesson, go back to building, and try again.
A real moment in Iteri
Screenshot of an actual Reading lesson will go here.
Image intentionally left blank
“The more I learn with Iteri, the more ideas I have for it.”
I started because I wanted a better way to study. Somewhere along the way, I found myself enjoying another kind of learning: figuring out how to make the thing I imagined, one imperfect version at a time.
← Back to my website