Code & AI

Selected work where design meets code and AI.

012026, ongoing

Designing the entry point for a self-service data & AI platform

What it is

A unified self-service entry point for data & AI platform interactions, bringing service discovery, access, provisioning, documentation, request tracking, and AI assistance into one experience.

What I'm doing

Developing AI Experience Design Framework to define what humans and AI should do, how they should work together, and where autonomy, control, verification, and recovery belong.

What's emerging

AI is changing how I design. Instead of designing and handing off, I'm foreseeing to work continuously with the people building the product and the intelligence behind it, learning what's possible, and shaping the experience together.

View the design framework

022026

Evaluated an AI assistant against its own design principles

What it was

An AI assistant added to Ericsson's internal HR platform, meant to help employees and line managers draft periodic performance goals faster.

What I did

Evaluated it two ways: scored every friction point against five published AI design principles (human in control, augment human capabilities, ethically aligned design, efficient automation, explainable AI) and ran a user study across people from various backgrounds and experience using it to generate real goals.

What I found

Explainability scored weakest of all five principles, and that gap showed up twice in the user study: nearly every participant rewrote the assistant's suggestions rather than trust them, and the more visible AI-assist entry point was used less than the plain 'Generate' button people defaulted to out of habit.

032025

Designed transparency into an AI-prioritized platform

What it was

A North American grocery distribution company's B2B messaging platform, routing urgent order and delivery updates to thousands of retail customers, was adding AI-driven message prioritization.

What I did

Designed a transparency framework around it: a visible explanation for every AI-flagged ranking, a feedback loop to correct it, and a manual override that never required going around the system. Research ran on two tracks: structured interviews, and AI topic modeling over a much larger forum-and-ticket corpus. AI had a hand in every stage of the design process itself, not just the feature, from that topic modeling in research, through drafting and stress-testing the transparency patterns, to reviewing the final designs.

What I found

The two tracks disagreed. Topic modeling surfaced a pain point interviews never once mentioned, operational bottlenecks customers had normalized as just how the process worked, and 8 of 8 users confirmed it as real once it was named back to them.

042023, ongoing

Generative studies

Short programs, mostly Processing. Some are a single output, some a set from changing the input, and a few respond to the mouse recorded.

#001

Generative composition #001: a grid of overlapping circles fading from blue to green left to right, colour driven by horizontal position

#001

//Dense pattern by embedding two for statements
size(500,500);
background(0);
noFill();
stroke(255);
for ( int i = 10; i < 500; i += 10)
  {
    for( int j = 5; j < 500; j += 10)
    {
      stroke(HSB,i,100,100);
      ellipse(i,j,20,20);
    }
  }

#002

Generative composition #002: a dense 8-petal flower silhouette formed from overlapping circles, offset 200, angle step 2 degrees

#002

//Composition with data generated using sin().
size(500,500);
background(0);
noFill();
stroke(255);
strokeWeight(2);
int oRadius = 250;
int offset = 100;
int iRadius = 300;
int modDeg = 15;
for(int deg = 0; deg < 360; deg += modDeg)
  {
  float x = oRadius + (sin(deg)*offset);
  float y = oRadius + (cos(deg)*offset);
  ellipse(x,y,iRadius,iRadius);
  }

#003

Generative composition #003: a dense red scribble of randomly placed triangles on black

#003

//Composition with random() function.
size(500,500);
noFill();
background(0);
for ( int i =0; i <500; i++)
  {
    int c = int(random(0,255));
    stroke(c,50,50);
    int x1 = int(random(500));
    int y1 = int(random(500));
    int x2 = int(random(500));
    int y2 = int(random(500));
    int x3 = int(random(500));
    int y3 = int(random(500));
    triangle(x1,y1,x2,y2,x3,y3);
  }

#004

Generative composition #004: a coarse vertical-streak texture from Perlin noise, step size 0.01

#004

//Creating textures with perlin noise
void setup(){
  size(1000,1000);
  noFill();
  background(0);
}
float xoff = 100;
float inc = 0.001;
void draw(){
  beginShape();
    for(int x = 0; x < width; x++){
      float y = map(noise(xoff),0,1,0,height);
      stroke(HSB,random(360),100,70);
      vertex(x,y);
      xoff += inc;
    }
  endShape();
}

#006

Generative composition #006: a large flower pattern of overlapping circles inside a zigzag ring border, iteration 250

#006

//Composition with rotate().
size(1000,1000);
background(255);
noFill();
strokeWeight(2);
translate(500,350);
int iteration = 250;
int c1 = int(random(125));
int c2 = int(random(175));
int c3 = int(random(10));
int c4 = int(random(50));
int c5 = int(random(255));
for(int x = 25; x < iteration; x++){
  int r = int(random(255));
  stroke(HSB,r,100,100);
  rotate(PI/x);
  ellipse(c1,c1,c2,c2);
  rotate(PI/2);
  rect(c3,c3,c4,c4);
  triangle(c2,c5,c5,c2,c5,c5);
}

#014

Generative composition #014: a black and white checkerboard grid

#014

//Coloring each pixel using C programming
#include <stdio.h>
int main()
{
    //pointer to the file and open
  FILE* fp = fopen("myfile.pbm", "w");
  // magic number
  fprintf(fp, "P1\n");
  // Size of the image
  fprintf(fp, "30 30\n");
  for (int i=0; i<30; i++)
  {
    for(int j=0; j<30; j++)
    {
      if( (i+j) % 2 == 0 ) {
        fprintf(fp, "%d ", (int)(1));
      }
    else if( (i+j) % 2 == 1) {
        fprintf(fp, "%d ", (int)(0));
    }
    }
    fprintf(fp, "\n");
  }
  fclose(fp);
  return 0;
}

052021

Ikat as a computational system

What it was

A 4-month, self-directed capstone: decoding the visual grammar of traditional Ikat weaving into a computational system, rather than reproducing its surface by hand.

What I did

Studied the weaving process at weaving centres before writing any code, reduced the pattern logic to 4 motifs across 3 grid structures, and wrote the rule set as a running 200-line program instead of hand-arranging convincing variations in a design tool.

What I found

The same motif × grid × colour logic that generated the original textile patterns generated interface and branding patterns just as well: the rule set was the reusable artefact, not the textile. It now runs a second time, ported to this site's own generative colour and pattern engine.

Read the full study