A mobile version while I was out biking with the kids
I spent an hour writing a plan, the agent worked for four hours — by evening the subtitle editor had a mobile version: video from the iPhone photo library, subtitles, burn-in and sharing the finished clip.
By Georg Malahov
On Saturday morning, May 9, I sat down and spent about an hour writing a plan: a mobile version of the subtitle editor. Not code — a plan, for ralphex. Ralphex is my harness around Claude, the AI assistant for code: you write a detailed plan, and the agent implements it itself, checks it and assembles the result. I kicked it off — and went biking with the kids, which I'd promised them the week before.
While we were riding thirteen kilometers there and back, with a playground and ice cream in the middle, ralphex worked for three to four hours. In the evening I opened the result — and I had a mobile version, a PWA: a web app you install on your phone. The whole flow works end to end: you load a video straight from the iPhone photo library in a few taps, create subtitles, burn them into the clip and immediately share the finished video to any other app. Plus a proxy for Deepgram — a speech-recognition service — and for subtitle cleanup and translation, so users don't need their own tokens like before.
There are rough edges: the quality of the subtitles themselves, quirks in editing, style presets that aren't very pretty. And one thing I consider critical: the preview doesn't always match the burn-in result. What's the point of a preview if the final video looks different? That one has to be fixed.
Where I actually got stuck was elsewhere. In a mobile editor there's essentially nowhere to put the video on screen: it makes more sense to hear what's being said and see the audio track. I wanted to extract the audio from the video and draw a waveform — so the phrases are visible to the eye and the subtitles move along them. I fiddled with it on Saturday evening after dropping the kids off, and half of Sunday on top — didn't work out. The task turned out to be hard and iOS-specific. I've set it aside for now, and I'm left with an open pull request — a batch of changes waiting to be merged into the project — of nearly a hundred commits.
The mobile version, by the way, I built on the new design system I'd made the day before with Claude Design — a tool for designing interfaces. What's left is to bring the desktop version and the presets in line with it, get the preview to match the burn-in — and since the backend proxy already exists, I can then add sign-in and let the first users through.
However much I tell myself there's little work left — no, there's still a lot. The new plan is this: within a week or two finish the blog and the Ukrainian association's site, put everything on the hub — and launch. Then I'll have a proper personal hub and at least three finished products — with a design that is decent by my own standards: one I actually like.