What stood in the way
Real algorithms—gradient descent, k-nearest neighbors, decision trees, k-means, neural networks and convolutions—had to run smoothly in the browser and stay mathematically honest while being friendly to first-timers.
Selected work
AI & ML / Case study
Model Visual
Six machine-learning models you can actually play with.
Model Visual makes machine learning feel tangible for beginners. Instead of formulas first, people turn the knobs, add data points and watch a model learn—no code and no sign-up.
Visit the live site (opens in a new tab)| Client | Model Visual |
|---|---|
| Role | Product Design & Interactive Engineering |
| Timeline | 7 weeks |
| Year | 2026 |
Delivery / 7 weeks
From first call to launch
- 01 · Discovery1 wk
Curriculum design and algorithm selection.
- 02 · Design1.5 wks
Visual language for data, models and controls.
- 03 · Build3.5 wks
In-browser model engines and SVG renderers.
- 04 · Launch1 wk
Math verification, usability testing and deploy.
How we moved forward
We built a React and Vite playground where every model trains on small data in real time, rendered as crisp SVG. Learners can click to add points, change the learning rate or noise, and step through training while live metrics update. Models are grouped into supervised, unsupervised and deep learning, with a guided learning path.
Delivered features
- Click-to-add data and live retraining
- Learning-rate, noise and step controls
- Supervised, unsupervised and deep-learning tracks
- Searchable model catalog and glossary
04 / Outcomes
What the launch delivered.
Outcomes come from the Nuoxsoft project record and the live site. They are not independently measured metrics.