Case Study: Eigodojo
How a two-person team took a product from research to high-fidelity prototype in two weeks without sacrificing quality.
The problem
It’s a very well documented fact that Japanese speakers struggle to learn English. This should not reflect poorly on them. You just have to watch an English speaker have a go at Mandarin, Japanese or any of the romance languages to see that this is a common issue. Speakers from different linguistic backgrounds all have languages that are less compatible with their mother tongue.
The data shows very clearly that English proficiency is declining in Japan and that adults have a very hard time sticking to their studies.
Another obvious data point is that apps are not helping. That comes up over and over on surveys. People using apps do not feel like they are making progress and they do not enjoy the experience.
The solution
Eigodojo takes a different approach to English study. Instead of grammar — which is already over-studied — we focus on high-frequency words and “chunks” built from common words, like “take it easy” or “sure thing”.
Short lessons that start by pre-teaching the necessary vocabulary and then a short dialogue where they can see those expressions used in a real context. After that, an easy quiz and then some feedback. Rinse, repeat and move on.
We use a spaced-repetition algorithm to score every expression the student encounters. That score controls the frequency in which they will be presented or tested on that particular word or expression. That keeps things moving and gives the student a clear idea of their progress.
Get Ricardo Costa’s stories in your inbox
Join Medium for free to get updates from this writer.
This approach is 100% based on research data. We are using a combination of methods that have shown actual results.
The Workflow
This project uses a heavily AI-supported version of Design Thinking. For each step we selected a set of AI tools that greatly improve our efficiency. This is an iterative, parallel, non-linear process. Every time we run through this loop we build more context for our agents, to a point where prompts get very short and the agents will surprise you by how much they know about the system and the research.
From Research to Ideation
The first three steps were condensed in an intense couple of days where we looked into hundreds of public sources and survey data produced by Learn Japanese Pod, observed themes and issues and ideated possible solutions. LLMs are excellent at quickly organizing ideas, spotting themes and producing useful summaries and reports.
Prototype
At this point, with a lot of the conceptual design work being done in parallel, we were able to skip all the low-fidelity and paper prototyping and go straight into medium fidelity (and soon high-fidelity) prototypes built in Sveltekit.
Claude Code became an essential tool. It allowed us to run through dozen of ideas and solutions with no friction until we defined the basic path for the product. This process also informed our design decisions and every component we made or tweaked was included in the design system.
Design
It’s hard to convey how much Figma MCP + Claude completely changed the way we work or even think about product design. Having a tool that can turn mockups into code, iterate on that code live and then plug the results back into Figma, ready for the next round, is just incredible.
This workflow removes most of the manual labor from the process. If you take your time at the start to define your visual style and have a solid set of rules and references, the agent will take it from there with very little guidance and give you consistent results every time.
This allowed us to spend time on the cast of characters and setting for the lessons. We decided against using Generative AI tools for those to maintain as much of their personality as possible. We still use AI tools to automate and optimize our 3D asset pipeline, but the creative process is kept completely in human hands, as it should.
The Results (so far)
At the end of the second week we have a mostly complete, high-quality design system and a good percentage of our brand guidelines are defined. We started on our cast of characters and set up a pipeline to produce the 3D assets and scenes we’ll need in the future.
We have an extremely high-quality prototype ready for user testing, built in a way that, even if we have to go back to the drawing board, we have an excellent toolkit to make the process painless and quick.
