Measuring how much text fits before the browser renders it.
To place text on a slide you have to know it fits before you draw it. Measuring with a canvas, wrapping Unicode correctly, and correcting for the gap between canvas and DOM.
Blog.
Notes on what we build, how we think about systems, and what we learn along the way.
Splitting text into balanced carousel slides without AI.
Where do the slide breaks go? A fixed character count cuts sentences and leaves uneven slides. How we turned the split into an optimization with a small cost model and dynamic programming.
Why HTML-to-image export uses the wrong font, and how HarfBuzz fixes it.
Rasterizing an SVG to PNG through an image element races the font load and usually loses, so the export comes out in a fallback font. The fix is to convert text to vector paths before rasterizing, with HarfBuzz for shaping.
Drawing connector lines between words in wrapping text.
An SVG overlay can connect words across two lines of HTML, but only if it knows where the browser put them. How we measure word positions and keep the curves attached through wraps, resizes, and font loads.
Storing your whole app state in a URL, with no backend.
A shareable link can carry the entire state of an app if you trim it, compress it, and version it. How we pack a full alignment into a few hundred characters, and read it back safely.
Splitting text into word boxes when splitting on spaces is not enough.
Every clickable word in the editor starts as a token, and a plain space split gets contractions, punctuation, and fixed phrases wrong. The tokenizer we built and the options it exposes.