Computer Vision in the Browser: Building the Laundry Label Decoder

Ever ruined a favorite cashmere sweater because you misread the laundry tag? You are definitely not alone.

Laundry care symbols are supposed to be a universal language, but in reality, they look like ancient hieroglyphics. A triangle with a line through it? A square with a circle and two dots? It is incredibly confusing, especially for expats and travelers dealing with foreign clothing brands. Our Development team set out to fix this universal, everyday problem using modern Vision APIs.

The Multi-Modal AI Approach

The Laundry Label Decoder relies directly on the newest generation of Vision-Capable Large Language Models, specifically OpenAI's GPT-4o architecture.

In the past, building an app to recognize symbols would require training a custom Convolutional Neural Network (CNN) from scratch. You would need thousands of labeled images of laundry tags, and maintaining the model would be expensive and technically exhausting.

Instead of taking the hard road, we leveraged multi-modal AI. We simply compress the user's photo directly in the browser and send the base64 string to the AI with a strict, heavily engineered classification prompt. The AI acts as the ultimate pattern matcher, instantly cross-referencing the symbols against global laundry standards.

Optimizing Image Uploads on the Client

A major technical challenge immediately presented itself during testing: sending raw 4K smartphone photos to the API takes too long and wastes massive amounts of bandwidth. Nobody wants to wait 15 seconds to find out if they can tumble dry their shirt.

To solve this, our Frontend Core Team implemented a highly efficient client-side canvas resizing utility.

// A simplified version of our client-side compression
const compressImage = async (file) => {
    // We draw the high-res image to an HTML5 canvas
    // Then we export it as a highly compressed WebP format
    // This reduces file size by up to 90% without losing the crisp edges of the symbols.
    return compressedWebP;
}

By compressing images down to under 200kb right on the user's device before the upload even begins, the API response time dropped from 8 seconds down to a snappy 2 seconds.

The result? A magical, frictionless experience where you snap a photo and instantly know whether you should tumble dry low or dry clean only.

Expanding Our K-Life Tools

The Laundry Decoder was our first foray into building highly practical, daily-use tools for foreigners living in Korea. It proved that AI doesn't just have to be for writing emails or generating code; it can solve tangible physical problems.

If you find yourself staring at a confusing clothing tag, simply open up the Laundry Decoder on your phone and let the AI do the work.

Following the success of this tool, we realized that navigating daily life in a new country presents many similar challenges. That is why we subsequently launched the Seoul Trash Map, another essential survival tool designed to help you quickly locate public disposal bins across the city. By combining AI vision and geographical data, we are slowly building the ultimate survival toolkit for expats.