Open Startup

Field notes

Welcome to the nerd zone! In our field notes we share how we build Reelang, let you peek inside the factory to see how the sausage is made, and get into how we use technology under the hood with our fellow geeks. Got feedback or ideas? Head over to our open roadmap and hash them out with us.

Just here to learn a language with videos, not the technical details of how we make it happen? Watch a video in the language you're learning.

August 4, 2026

In l10n, an LLM with context beats a human translator without it

If you are a small team shipping in more than one language, you can get native-quality localization out of AI models today. Actually good, the kind that survives a native speaker reading it closely. What decides the outcome is not which model you pick, but the context that reaches whoever is translating, whether that's a model or a human. The diligence of models can have an advantage over a professional you would hire otherwise.

Read

August 3, 2026

Portuguese speaks two ways

Brazilian and European Portuguese share a name and five centuries of drifting apart. A look at diglossia in Portuguese, the words that mean two different things depending which coast you're on, and what it takes to measure "common" once you split it.

Read

July 22, 2026

How to build a Simple English version of your website with LLMs and word-frequency data

If your site has a global audience, some of your readers are not fluent in the language you wrote it in. A proper localized site is the best answer, but there is always a long tail of languages you can't translate and maintain. Here is how we build a plain, Simple English version instead: quickly, with reasoning LLMs, and grounded in real word-frequency data so it's actually readable.

Read

July 8, 2026

Hiding the words so you can hear them (blurring out burned-in subtitles for language learning)

Half the videos worth learning from arrive with subtitles stamped right into the picture. Great for scrolling with the sound off, quietly bad for your ears. Reading a line before you've heard it isn't the same as understanding the audio. So we built a way to find those baked-in subtitles and blur them out, and paired it with our own tappable subtitle layer that you actually learn from. Here's the thinking, and, in plain terms, what it takes.

Read

July 1, 2026

How we find videos you can almost understand

Nobody tags a YouTube clip "good for a high-A2 learner of Portuguese." So we built a pipeline that goes looking, in every language, and grades each video by ear before it ever reaches you. Here's roughly how it works.

Read

June 27, 2026

Why an accurate video transcript is the foundation of Reelang, and why we run it on Deepgram

Almost everything clever Reelang does, the smart subtitles that surface the most common words, the difficulty scoring, the learner insights, rests on one plain thing: knowing exactly what was said, and when. That is a speech-to-text problem, and it is the part of our stack we worry about least. Why we run it on Deepgram, which models we use per language, and how the word timing drives everything.

Read