Build notes
Building AiNerd.sg with Astro, GSAP and Lenis
Inside the first experiment: give a static field journal a physical character, then let motion explain the layers.
The question behind this site was whether a static journal could feel like a small physical place. The robot had to carry the first impression. A person arriving to read about AI also needed a direct route from that first screen to a useful tool.
Give the mascot a material
Nerd-01 began as an original AI-generated studio portrait. Light catches the edges of the ivory shell and the metal around its joints. The dark glasses frame two bright pixel eyes. Those details still describe a character when the image is small, or when motion is switched off.
In the first version, the whole image shifted under the pointer. The eyes stayed fixed inside the glasses, which made the robot feel like a photograph being moved around. For the second version we separated the head from the torso, extracted the cyan eye pixels and filled the areas they used to occupy. Now the eyes can glance toward a word while the head gently tilts and the shoulders stay put.
Try moving slowly from the title to the robot, then tap its face. The little Singapore flag is attached just below the antenna ball. Its free edge ripples while its hoist stays fixed. The glasses still have one photographed viewpoint, so the head turns only a few degrees.
Put the movement on one clock
A short timeline introduces the title in two beats. The tool shelf follows in ordinary document flow. Further down the page, a set of elliptical rings rotates as the reader scrolls through the experiment. Changing the angle of each ring creates the impression of a wire sculpture, while the background grid travels a different distance.
On a fine pointer, Lenis supplies wheel smoothing. GSAP advances Lenis on its own ticker and ScrollTrigger receives the scroll updates. Touch readers use native scrolling. A reduced-motion preference keeps the sculpture in a fixed pose. Try toggling that preference, then compare the scene at the same place on the page.
What counts as success?
Start with a practical test. Can you follow Discover, open a tool and read a whole journal entry with JavaScript turned off? Does the mobile menu give keyboard focus back when dismissed? Does the portrait reserve its space before the image arrives? These checks describe different ways a beautiful introduction can interfere with reading.
Visual polish is a judgement call. The more repeatable part of this experiment is the division of labour: HTML carries the content, CSS establishes materials and spacing, and a small set of motions gives the robot a presence. The next experiment can change the motion without rewriting the journal.
What the browser check found
In the local production preview, the journal index and article links worked with JavaScript disabled. With reduced motion enabled, a pointer crossing the portrait left its transform unchanged. In the first build, a desktop pointer shifted the portrait, the tool cards lifted, and wheel scrolling advanced the page. These observations tell us which interactions survived the build.
The first narrow-screen capture also showed the field note and its link competing for a single row. Giving the link its own line let the note read as a sentence again. That small layout change mattered more on a phone than adding another animation.