Fixing a Homepage After 3 Years
My wife currently runs an unmanned cafe. And three years ago, I built the cafe’s homepage on WordPress.
Since then, I’d barely touched it. Menus changed, images changed, but the homepage just sat there. Maintaining it was too much of a hassle. One day I opened it up and it was completely out of sync with reality — discontinued items were still listed, and current items were missing.
That wasn’t going to fly anymore. So I decided to renew it. (Actually, I’d been meaning to since last year…)
This time I decided to rebuild it from scratch with a structure that would be easy to maintain. And along the way, I decided to lean heavily on AI.
Challenge 1: Sorting Out 90 Menu Images
The first wall showed up as soon as I started the renewal.
There were 90 menu image files (just for the grind-type machine). All the filenames were in Korean, and the sizes were all over the place. To put them on the web, I needed English filenames, and they needed resizing too. Opening 90 files one by one, renaming them, resizing them — just thinking about it was overwhelming. And one mistake in the middle meant starting over.
I asked the AI:
“Make me a mapping table from Korean filenames to English ones. All 90.”
The AI read each menu name and built a mapping table with English-style slugs.
Then, based on that table, it wrote a Python script that handled resizing and renaming in one pass. I ran the script, and all 90 files were sorted out at once. I didn’t even have to open the images individually. What would’ve taken half a day by hand — no, probably a full week — was done in a few exchanges.
Challenge 2: Rebuilding the Menu Data from Scratch
Once the images were sorted, the next problem showed up: the menu data. The menu had changed a lot over three years. Some items were new, some were gone. I had to rebuild the names, prices, categories, and temperature options for 25 drinks from scratch.
But the current menu was already fully displayed on the kiosk screen.
I showed the AI a photo of the kiosk.
“Based on this photo, turn the menu into JSON data.”
The AI analyzed the photo, sorted the 25 drinks into categories, and extracted structured data.
- 4 Americano-family drinks
- 5 Cafe Latte-family drinks
- 5 Specialty Lattes
- 5 Milk Teas
- 6 Ades & Teas
It commented out items only available on a specific machine, and stripped out details unnecessary for the web, like origin labeling. Hot/iced distinctions were sorted too. I just dropped a single photo into the chat, and the menu data was done.
Challenge 3: How Do You Describe the Flavor of a Non-Coffee Drink?
Once the menu data existed, I had to figure out how to describe the flavor of each drink. The old homepage used the same flavor axes for every drink: body, acidity, bitterness.
But a bitterness gauge on a strawberry latte is awkward. Body on a grapefruit ade? That was a problem I’d just glossed over when I first built the homepage three years ago.
This time, I decided to get it right. I asked the AI:
“What flavor axes would fit non-coffee drinks?”
The AI analyzed the characteristics of fruit drinks, milk teas, and latte-family drinks, and proposed alternatives. The answer was sweetness, acidity, refreshment.
We also settled on the classification rules together.
| Condition | Category | Flavor axes |
|---|---|---|
| Name contains “Cafe” | Coffee | Body / Acidity / Bitterness |
| Contains only “Latte” | Not coffee | Sweetness / Acidity / Refreshment |
| Choco Latte, Double Choco, Caramel Latte | Latte without coffee | Sweetness / Acidity / Refreshment |
The whole rule was settled with one line: “a plain ’latte’ just means it doesn’t have coffee in it.” A problem I’d glossed over three years ago got resolved in a single conversation.
Challenge 4: Finding the Right Voice for an Unmanned Cafe
Once the flavor system was in place, I had to write intro copy for each drink.
The old homepage had lines like this:
“A body-forward latte, gently foamed with steam”
Cafe Purple Mint is an unmanned vending machine. There’s no steam. There’s no barista. For three years, we’d been running an inaccurate description.
I explained the situation to the AI:
“Our cafe is an unmanned vending machine. A machine makes the drinks, no barista involved. Write copy that keeps that in mind while still making the drinks sound appealing.”
Here’s what the AI rewrote it as:
“Precision-extracted espresso meets fresh milk for a deep, smooth latte, consistently delivered by machine every time.”
The unmanned aspect turned into a strength instead of a weakness. That one word, “consistently,” turned into a sense of trust — the same taste, every single time. I described the cafe’s character, and the AI found the language that matched it.
Challenge 5: This Time, Building for Easy Maintenance
The biggest goal of this renewal was maintainability. The reason I’d left it untouched for three years was that it was “too much of a hassle.” (Try WordPress sometime. It’s hell. Absolute hell.)
While building the UI with the AI, I focused first on a structure that would make it easy to add and edit menu items going forward. Along the way, a bunch of small and not-so-small problems came up.
Too many items, so the page scrolls forever.
→ Switched to a sliding carousel showing 5 items at a time, auto-advancing every 5 seconds.
Can’t tell an ade from a milk tea at a glance.
→ Analyzed the naming patterns and added automatic badges.
`Grapefruit Ade` gets an Ade badge, `Earl Grey Milk Tea` gets a Milk Tea badge.
Images get cropped top and bottom in the modal.
→ Changed `object-fit: cover` to `object-fit: contain`.
#### The flavor gauge doesn't render properly on mobile.
→ The culprit was `align-items: flex-start`. Fixed it to `stretch`.
Whenever I described a problem, the AI pinpointed the cause and fixed the code. Each time a bug got squashed, the page got a little more finished.
Conclusion: I Built a Homepage Through Conversation
The reason it sat untouched for three years was simple: it was just too much manual work.
This renewal was different. Batch-processing 90 images, generating menu data from a single kiosk photo, designing a flavor system, writing copy, building the UI — all of it was worked out by talking with AI.
The AI didn’t decide the direction. The person who knows the cafe best set the direction, and the AI quickly drafted along that direction. If I’d been on my own, I don’t think I would’ve even attempted it. Or rather, I probably would’ve put it off for another three years.
Give it a try. It’s a lower bar than you’d think — anyone can clear it.