an image

Speak with AI in Japanese

Acorn Japan

I started working on a Japanese language learning application focused on helping learners practice speaking Japanese out loud.

Recently I spent a week or so in Japan. It was the first time I'd been there since studying as an exchange student 10 years ago.

I do not know very much Japanese. I would like to know more, but studying is boring.

But trying to talk to people is pretty fun. I don't mind it. Reading and writing is much less interesting. Listening to pre-written scripts and trying to understand those is no fun. Private tutoring is expensive. I'd practice with my wife and relatives but it's tiring for them to explain everything to me (over and over).

I think AI models might be in the exact right state of development that they can seriously help with this. I want to try.

The first week

Wordpress homepage

I bought acornjp.com and set up a basic wordpress site as a landing page. I created a cute logo, some app mockups, and other media with GIMP and MidJourney.

I want to avoid going into the cave and building without trying to reach an audience, so I started here. There are already several changes I want to make to it.

Proof-of-concept

Dom helped me write a proof-of-concept of

These steps were not connected together; I manually copied some data or files from one to another. Connecting them all is on my TODO list.

I confirmed that the data flow works in both English and Japanese.

Cloud setup

This step has taken the most time, because I (foolishly?) decided to try a bunch of tools I'm not familiar with. Instead of just Kubernetes (which I'm already only kind of familiar with) running various containers (including postgres), I am also using all of these services for the first time: helm, terraform, google secrets manager, google cloud sql. Getting a basic cluster up and running has taken MUCH longer than it needed to, and this isn't the "interesting" part of my idea. So it may prove to be a waste of time, especially because I'm racing against the clock trying to leverage the initial burst of motivation (which tends to wane within ~ 14 days) to reach an demo that's interesting enough to show to people.

Project organization

I ended up creating 10 git repos: One meta repo and 9 git submodules. I'm not sure whether this was a good or bad idea; I just didn't like the idea of having everything combined into one. But the number of repos exploded rather quickly, mostly because each of the components in the stack (hub, stt, tts, chat, browser-client) have their own repos, and there's also ops, records, and media. (Ah, so maybe I created 8 submodules, not 9).

Genki Volume 1

I bought the Japanese language textbook and workbook that I used in school when I was studying Japanese. Hopefully this will help me figure out good prompts for the demo.

Next week

I'm not quite at the end of week 1. But I suspect the next few days will be a bit busier at my day job and it'll be harder for me to make progress, so I'll list some of the things I'm thinking of doing next:

After next week

The idea for this project excited me because there's an enormous number of things I'd like to try to do in the space, which means a couple things. It means that it's fun to let my mind wander and imagine the many possibilities. It also means that prioritization, design thinking, and being opinionated about what I should and should not try to do is very important.

This project is in the "let's build something basic, try to put it in front a target audience, and see what happens" phase. If "what happens" is "they're really excited about it (or the idea of it)" then I may turn it into a bigger focus.