Notes on the last few months of changes and UI experiments for my personal app Japan stock screening (yfinance × Investment Strategy).
How it’s built and how data is fetched is already covered elsewhere—see the links below if you want the full story.
This post is about recent updates and what I tried with Claude and DESIGN.md.
Earlier write-ups
- Zenn — Screening 3,700+ Japanese stocks with yfinance and Investment Strategy (vibe coding, in Japanese)
- note — Screening almost all listed names + US stocks from ~13k tickers (Japanese)
What changed
Since then I shipped:
- CSV kept in the browser: After a CSV loads once, the app remembers it so you can keep screening on the same dataset without reloading every visit. Small quality-of-life change—I didn’t love losing the data on each session.
- Search presets: You can save and apply filters you use often as presets.
- Screener UX/UI: Tweaks around readability and day-to-day use—layout, density, and small interaction details.
Visually it looks roughly like the captures below (before on top, after on bottom). The after state reflects Dataset view, a favorites tab, column toggles, and colors/layout reworked to fit the new behavior.
Asking Claude for a “more modern, screener-friendly UI”
I wanted the UI to feel one step more modern, so I tried a prompt like this in Claude:
Make this design feel more modern and easier to use for screening
(target URL:https://yfinance-jp-screener-search.vercel.app/)
The responses looked clean and promising.
I mostly treated it as reference only for now.
If I do a full redesign pass, I’d probably reach for Google Stitch first.
Applying a Wise-style DESIGN.md
Next I tried DESIGN.md from Google Labs’ Stitch.
getdesign.md hosts lots of service-style samples.
There’s also a Japanese-oriented collection here (handy):
I grabbed the Wise design-md (Wise / design-md) because I liked the tone, dropped it into the repo as DESIGN.md, and asked Claude or Cursor to remap the current UI to that spec. That was basically the whole workflow.
Buttons and palette skewed Wise-like and closer to my taste, but it didn’t feel like a big leap overall—so this one is on the shelf for now too.
Closing thoughts
The web side of things seems headed toward dropping a small declaration file (DESIGN.md) and remixing layouts from it.
I can imagine the same pattern—text-first specs driving implementation—spreading across other fields the way games already lean on data-driven pipelines.
It feels like people who can go from zero to one keep getting stronger leverage.
- Broader vibe-coding context: Vibe coding: three shipped projects as a web beginner
- When GitHub Actions artifact storage blew up: GitHub Actions artifact storage quota (free plan)
