I run my own website the way I tell clients to run theirs. This week that meant a hard audit, and the result was not the one I expected. The engineering was fine. The problem was the first thing anyone sees.
The site is abhirajdg.in, my personal one. It has four pages for four sides of me, each its own world, and I wanted it to meet the standard a design jury would hold it to: design, usability, creativity and content, plus a technical gate for speed and accessibility. So an AI agent audited it against that scorecard, in a real browser, route by route. It is the same discipline as when we scored adg-advisory.com before redesigning it, pointed at a different site.
What the audit found
Most of it was housekeeping, and housekeeping is where the honest wins hide. The build was clean. Nothing scrolled sideways on a phone. The 404 page worked. The list of things to fix was long and small:
- On a phone, 22 of the 59 tappable elements on the home page were smaller than the 44 pixels a thumb needs.
- There was no skip link, so a keyboard user tabbed through the whole menu on every inner page.
- The keyboard focus ring was being cut off on the site's angular buttons and cards, so a keyboard user could not see where they were. The code looked fine. It only showed up in the screenshots.
- The error page played a sound even when the visitor had turned sound off, and it forced a redirect with no way to pause it.
- The framework was behind on security fixes, including critical ones.
We fixed most of it in one pull request. It is the right place to start and the wrong place to stop, because none of it changes what a stranger feels in the first three seconds.
The bottleneck sat upstream
This is the idea behind the Hexagram Audit: when several things look wrong, one of them is upstream of the rest, and fixing it first changes everything below it. That is what the Architect pillar is for. Here the upstream point was the home page's first screen. It opened with a small portrait, a name and a paragraph. The idea that makes the site interesting, that one person has four lives and you pick which one to meet, sat in a card grid underneath all that.
My read, and it is a judgement rather than a measurement: the audit's real finding was a positioning problem, not a technical one. The concept was strong and the first screen hid it. We had polished the rooms and left the door plain. If that sounds like a problem you have, it is the kind of thing brand positioning work exists to fix.
What we built instead
We made the idea the whole first screen. It is now a character select, like the one on an old arcade cabinet. You move between four characters with the arrow keys, a gamepad, a mouse or a tap, and each one decodes its own one-line description from a bit of machine code into plain English. Pressing start switches the screen off like an old television and opens the page you chose. That decode is also the plainest description of my job I have found: translating between humans and computers.
Two choices kept it honest:
- Everything is still a real link. With scripting off, the four characters are ordinary links and the text is all there. With reduced motion on, the screen is still and navigation is instant.
- The plain answer stayed. The sentence about who I am and what I do is still in the first section, as ordinary text outside the animation. Everything that used to be on the page is still there, further down.
Where it went wrong
This is the part I would rather skip and should not.
Two fixes made things worse, so we reverted them. Mobile load speed was the one gate we did not close: on a local test build the main content took between 2.3 and 3 seconds to appear, against a target of 1.8. We tried loading fewer fonts up front, which helped one number and hurt two others, including a layout shift on one page. We tried inlining the stylesheet, which scored worse. Both are reverted, and the gap is still open.
The redesign did not go live when I thought it had. It was built on top of the fixes, and I merged both within a minute of each other, before GitHub had re-pointed the second one at the main branch. The new home page landed in a side branch instead, and I only noticed when I looked at the live site and nothing had changed. It was re-landed and went live the next morning.
What a founder can take from this
Audit the whole thing before you touch any of it, and score every part against the same list. Fix the housekeeping, because it is cheap and it is real. Then find the one upstream point that decides how everything else is received, and rebuild that.
Every change reached the live site as a pull request that I merged. The agent did the measuring and the building; I decided what shipped.
The code was fine. The door was the problem, and nobody sees the code.
If your marketing has the same shape, with plenty of effort in the rooms and a plain door, the Hexagram Diagnostic is the free first step.
