A small studio in Seoul

Web, apps, and the AI that works inside them.

We design and build websites and app services. We also build and run our own AI tools — ones that measure how often a brand appears in AI answers, check content before it ships, and find things across text, images and video.

See what we built contact@unipurple.co.kr
Illustrative run · brand and numbers are examples
What we believe

Most AI tools summarize. Ours are built to measure, check, and leave a record — so a person can see why something passed, and decide.

The difference matters when the output goes out under your name.
Work

Tools we built and use every week.

Each of these runs on real projects today. Where we show a number, we say whether we measured it or it's an example.

AI Search Visibility Meter

Asks ChatGPT, Perplexity, Gemini and others the questions real customers ask, and measures how often a brand shows up in the answer. Judgment combines a local model with fixed rules, so the same answer always gets the same score.

ChatGPT
47%
Perplexity
38%
Gemini
26%
Claude
33%
Copilot
17%
beforeafter the fix

We don't stop at the diagnosis. We change the content, then measure again.

Illustrative values · measured twice

Pre-publish Review Gate

Reads a draft in context, not against a word list. Flags risky phrasing and checks every claim against a source.

Our clinic offers a guaranteed result with a recovery time of 3–5 days, as noted in the doctor's guide.
needs reviewsource matchedhuman approves
Illustrative draft

Visual Production Pipeline

Images and video made against an approved visual spec, then checked — so quality repeats.

1
Visual contractwhat must and must not appear
2
Compiled promptbuilt from the contract, not free-hand
3
Generateimage or video models
✓
Proof checksinspected before it's used

Multimodal Search

Puts documents, photos and video in one space so you can find them by describing them. It runs on the device, so files never leave it.

"close-up of a model with braces"
0.81nDCG@10
Korean retrieval benchmark, 300 queries. The model we used before: 0.77.
5/5
Videos ranked in the top 5 from a one-line Korean description, among 184 assets.
Measured by us · October 2026
How we work

Measure. Change. Measure again.

A diagnosis alone doesn't help anyone. Every project closes the loop with a second measurement.

1

Measure

Find out where things actually stand — visibility, quality, speed — with numbers we can repeat.

baseline
2

Change

Fix what the numbers point to: the page, the content, the workflow, or the tool itself.

ship
3

Measure again

Run the same measurement and compare. If it didn't move, we say so and go back.

verify
↺ the third step becomes the next first step
Principles
AI decides.
AI finds. People decide.
evidence

Measure before we claim

"It got better" needs a before and an after. We keep both.

judgment

A person makes the call

Our tools point to what's risky and why. Approval — and the record of who approved — stays with people.

privacy

Your files stay inside

Where the work can run on the device, it does, so client material doesn't travel.

Services

What you can hire us for.

Small team, direct line. You talk to the people who build it.

Websites & landing pages
Planning, design, build and launch — structured so both search engines and AI assistants can read it.design · build · deploy
App services
Web-based products and internal tools, from first sketch to running service.product · engineering
AI search visibility
Measure how often you appear in AI answers, fix the gaps, and measure again.diagnose · improve · re-measure
AI tools in your workflow
Review gates, multimodal search and production automation, fitted to how your team already works.integration · training

Have something to build?

Projects, partnerships, or just a question — write to us.

contact@unipurple.co.kr We reply in English or Korean.