t0kt0

writing about web craft, CSS, design engineering 🌈 by & for diversity in tech

1 September 2023

careervuecanvas

BYBORRE: from pixel to needle

Building a browser editor for knitted textiles, and the tests, releases and backups that made it safe to keep shipping.

At BYBORRE from August 2021 to 2024.

In August 2021 I joined BYBORRE, a textile company in Amsterdam. I started remotely from Buenos Aires, Argentina, on a short freelance contract, and a few months later I flew to Amsterdam to stay. I didn't know anything about industrial knitting. I had, however, knitted a sweater using circular needles. Three years later I could explain why big letters are a bad idea in a 3D textile.

What Create does

The Create dashboard today: customize a textile, or design one from scratch.

BYBORRE designs and knits its own textiles. BYBORRE Createā„¢ is the web app where designers and product developers design and order custom knitted textiles. At its center is a textile editor built with Vue and HTML canvas, running on AWS.

The idea behind it is simple to say and hard to build: pixel to needle. Every pixel in your design maps to a needle on the knitting machine. While you design, the knitting file gets built. Before that, knit programmers did it by hand.

Translating knit into pixels

The Create editor in 2023, with the fill pattern picker open. Image: BYBORRE press kit.

From 2021 to 2023, a lot of my time went into making the canvas behave like a textile.

  • Fill patterns (September 2021). A knitted textile has a front and a back layer, joined at attachment points. The editor reads the canvas pixel by pixel, finds the attachment pixels, and swaps in the right back-layer yarn colors for the colors the user has active. Colors on a canvas are never exactly what you set, so I added a small tolerance per RGB channel. Without it, a pixel that was off by two values broke the pattern.
  • Text and outlines (December 2021). The knit experts in our lab taught me that a big letter with a continuous attachment line puts too much tension on the fabric. So text and outlines got rules: minimum and maximum sizes, spacing thresholds, and only on the textile types we had validated.
  • Color (early 2022). A color on screen is not a yarn. I researched color matching (Delta E in the CIELAB space) to map what users picked to the yarns we actually had, implemented multitones, and wrote it all down in the wiki.
  • Performance (2023). The editor handles big images and big samples. I cut loading time in the sample list and the editor, and I refactored saving so we save one sample at a time. That reduced waiting time, and it also stopped people from losing work.

An overview of the editor: pick a textile, upload artwork, split it into layers, apply a fill pattern and yarn colors, and save.

The best part was learning from the knit lab. I spent time with the knit programmers to understand how a design goes into the knitting machine software, and used that to improve the files we generated. It felt like learning a new language, and I love learning languages.

Making it safe to keep building

When I joined, there were no tests, no written release process, and no backups we could trust. That is normal for an early product. It just doesn't scale.

So, step by step, starting in September 2021:

  • I wrote the first automated tests, for login and for creating and saving a sample, then mapped the critical flows and wrote a full test-case suite for the rest: user roles, the admin, the textile catalog, the shopping cart and the editor.
  • I documented the release: development, then staging, then a production database backup, then the yarn and color migrations, then production. Later I automated most of it and added tagged releases.
  • We separated the testing environment from production completely.
  • In April 2023, I scripted the database backups and set up a local database in Docker, so nobody had to test against real data.
  • I reviewed the repo and proposed the basics: shorter PRs, lint hooks before commits, a clear store structure, and a style guide.

I wrote this in a sprint demo, and I still believe it: software was written and maintained by human beings (before AI), and as humans we might forget how we do things or why. Digital products often come without a manual, so someone has to write it.

This work mattered when the company raised its Series B. In February 2023, BYBORRE raised €16.9M, led by Invest-NL and VP Capital, to grow internationally and keep building the platform and the design tool. Before that, the investors ran a technical due diligence on Create. They asked about tests, security, infrastructure and package licenses. A year and a half earlier we would have had very few good answers. This time we had them, plus a clear list of what to fix next, and the product I had been building was part of what the company was raising money for.

The configurator

The editor is powerful, but our users told us it was too much. Architects and interior designers didn't want to learn an editor. My colleague Luisina Salce, our product designer, summed up her research in one line: "Customers want uniqueness, without the complexity."

So in 2023, together with textile engineers, designers and the dev team, we brought customization to the product page. You pick a BYBORRE textile, then change three things: color, shade (three levels) and repeat size (three sizes). That's more than 500 variations without opening the editor. The hardest part was rendering: showing depth, shadow and the feel of the material in the browser was more complex than any of us expected.

The configurator on the Eva product page: color, tone and scale, rendered on a 3D sofa.

What I learned

When I interviewed at BYBORRE I said contributing to a codebase was what I loved most, and I wasn't keen on managing a team. I still love the code. But somewhere between the knit lab and the due diligence calls, I found I also like building bridges: taking a complex engineering problem and explaining it so a textile designer or an investor gets it.