Cognisian Website
The company's own site, seventeen pages of it. A healthcare data and AI company selling to hospitals and clinics, which meant deciding what the company sounds like before deciding what it looks like. I designed and built the whole thing.

The Problem
A healthcare intelligence platform is a hard thing to put on a page. The product connects clinical, operational and financial data, which is abstract, and the people buying it are hospital and clinic executives who have been sold dashboards before.
So the site has two jobs that pull against each other. It has to feel like a company you would trust with patient data, which argues for restraint. And it has to be memorable enough that someone remembers the name a week after a conference, which argues for the opposite.
The Decisions
- 01
Three doors, not one
There isn't a single solutions page. There are three, one each for hospitals, clinics and healthcare groups. A hospital and a two-room clinic have almost nothing in common in what they're worried about or what they can spend, and a page written for both says nothing to either. Three pages is three times the copy to keep current, which is the cost. It also means a visitor sees their own problem described rather than a category they have to translate themselves.
- 02
Marketing can publish without me
News and case studies are MDX files with front matter, not entries in a CMS and not hard-coded pages. If publishing a post needs a developer, posts stop happening, and a company site with a news section last updated eighteen months ago is worse than one with no news section. This way a new post is a file, and I'm not the bottleneck.
- 03
The demo request is the only thing that costs money if it breaks
It's the one route on the site with its own test file. The form validates against a schema on the server as well as in the browser, it's rate-limited before it reaches the mail service, and the mail goes through a transactional provider rather than SMTP. Everything else on the site is words. This is the part where a real person is trying to reach a salesperson, so it's the part that gets the belt and braces.
- 04
Accessibility is a test, not an intention
The end-to-end suite runs axe against the pages, and the linter carries the JSX accessibility rules, so a missing label or an unreachable control fails the build rather than waiting for someone to notice. Healthcare buyers include procurement teams who ask about this in writing. Beyond that, saying a site is accessible and testing that it is are different claims, and only one of them survives contact with a redesign.
What Shipped
17 pages, from the home page to three pages written for hospitals, clinics and healthcare groups.
Radix for the hard parts
Menus, dialogs and selects are Radix, because they're hard to get right by hand. Content that changes often is MDX.
An opening that assembles
A WebGL particle field forms the word REIMAGINE, drawn without a 3D library.
The demo form can't quietly break
Validated, rate-limited, and tested on its own.
Tested
Unit, end-to-end and accessibility checks. 122 commits.
- Next.js
- React
- TypeScript
- Tailwind
- Radix UI
- MDX
- OGL
- Zod
- Playwright
- Next.js
- React
- TypeScript
- Tailwind
- Radix UI
- MDX
- OGL
- Zod
- Playwright
Where It Is Now
Live at cognisian.ai, and the company's front door. Last commit was 2026-07-20.
The honest limit on this one: I can tell you what it does and how it's built, but not what it converts. Enquiries land with the sales side, and I haven't closed that loop back to the page that produced them. On the eye-surgery landing page I did, and it changed what got built. This one is still a site rather than a measured funnel.

