Load it, close the door, done.

A washing machine cleans reliably these days. The complicated part is getting there: choosing the right program. Modern machines have sixteen to twenty of them, with names like "Eco 40-60" that almost nobody can decode. So most people stick to the same two or three and leave the rest untouched.

The strange thing is that the hardware is already smart enough. Many machines already measure load size, soiling, and fabric type, and dose the detergent automatically. What's missing isn't the sensors, it's the layer above them that turns this knowledge into a decision you can follow. The machine has grown smart at carrying out a wash, yet stayed blind to what it is actually washing.

That was the question driving this project: what does the interface of a washing machine look like when the machine itself holds the knowledge about fabrics and programs, and brings you along instead of taking over?

I started from a single assumption: a camera with a macro lens sits in the loading area and recognizes each item as it goes in, its material and color. I sketched the flow on paper first, the way I almost always work, then built a clickable base in Figma and turned it into a plain UI. The central rule emerged quickly: the most delicate item sets the terms for the whole wash. A wool sweater in a load of cotton pulls everything down to 30 degrees, because wool felts otherwise, and that can't be undone.

How the system decides runs across two separate layers. Perception recognizes material, color, and soiling through the camera and works, as neural networks do, with a degree of uncertainty. The decision, by contrast, follows fixed rules rather than a learned model, so it stays verifiable. It runs in two steps: first the most delicate item sets the hard limits for temperature and spin speed, derived from the care symbols under ISO 3758. Within this permitted range, the system then optimizes for low wear and energy use. Only one decision is left open, the one no sensor can make: when the laundry should be done. The diagram shows this sequence using a wool sweater as the example.

What came out of it is a design concept and an interactive reasoning explorer. Laundry items are placed into a simulated drum, and the temperature, spin speed, program, and the reasons behind them change live as the load changes.

Three cases cover the logic. A cotton load runs at 40 degrees; adding a wool sweater drops it to 30 degrees and 600 rpm, with the line "without it, 40 degrees would be possible"; a black sock leaves the temperature where it is and instead triggers a warning to separate it out. The concept and prototype came together in a single workday, the UI in Figma, the explorer as HTML with simulated recognition.

The camera recognition has its limits. Throwing three items in at once can overwhelm it, which calls for re-classification with a visible confidence status rather than a silent assumption, and the same goes for washed-out or missing care labels. More complex use cases are still open, like setting a specific finish time, delaying the start, or washing sneakers, and a camera plus the computing power in every drum raises the unit cost.

The UI is at concept stage too, not yet worked out in detail; there wasn't time here for a finished visual design. It also remains open whether the machine should learn from my habits. My current position is no, because reliability comes before personalization.