Quick start
Listen Continuous mix at your level Study Words and common phrases Reference Kana, particles, verb forms Practice Dialogues in different situations
Schematic map of Kyoto School — grammar revision University — advanced patterns Journeys — take a walk through the city
Listen Continuous mix at your level Study Words and common phrases Reference Kana, particles, verb forms Practice Dialogues in different situations

8,001 words and patterns, 60,605 example sentences, 550 short and 24 long dialogues, 74 hours of Japanese audio

What is Geeks in Kyoto?

We're the course for people who want to spend time in Japan:

Our course is for self-powered adult learners who are interested in communication. The material is mostly spoken, usable on your phone and even offline. We're not a reading course, but there's an 'interactive book' mode for when you're on a laptop: read fully annotated dialogues, see how sentences are built up, and type your own lines — the AI corrects you and replies in character.

We're in development and not ready to take payment yet. Right now you can browse the whole course, and you can make a free account for roughly half an hour of listen time per day. Get unlimited when paid plans become available.

Our plans (indicative while we build) →

How it's made

Geeks in Kyoto is original content, born of the frustration of studying for years and being unable to catch natural Japanese at nomikai. That's what the course exists to fix. It's generated by AI and authored, cast, and checked by me.

I write the scenarios in the case of longer dialogues, or the briefs for shorter examples. I set the tone of what to teach, what to voice and how to cast it, and what register of Japanese to use in each situation (casual, polite, or a mix). My personal anecdotes from travelling in Japan over 20 years are woven into the stories.

Claude AI does the site coding, expands a scenario into beat-by-beat who-did-what, and together we adjust details of the casting or the brief. Gemini AI then generates the Japanese text as well as the voices. Every voice is synthetic. Our characters are fictional and casting is by general description ("an older, educated man") not voice cloning.

Both Claude and Gemini produce natural Japanese, so potential errors are rarely in the grammar. They're in AI 'hallucinations', continuity and script-following, kanji readings, situational awareness, and tone.

To guard against these, first we automatically transcribe the audio content and check that it matches the script — that's necessary as the current state of AI makes unpredictable mistakes. Claude AI does a critical review of the Japanese to spot mistakes, and I listen to the whole course to see if it feels right, appropriate, and logically together. At the end of the day I — Pavlos Papageorgiou via Geeks in Cities Ltd (SC885112) — hold editorial responsibility for everything this course publishes.

If you do find something wrong please report it with the 'Report AI content' button and we'll fix it right away.

The full story — pipeline, review process, and editorial responsibility →