GPT-4o built a braille-friendly lyrics app in 3 prompts

lines of HTML codes

Developer Ashley had a specific problem: lyrics on the web are poorly formatted for line-by-line scrolling on a braille display, and navigating most lyrics apps as someone less proficient in braille is a friction point. So Ashley built a native app to fix it, using hands-off vibecoding for the first time.

The result is Lyrx, an open source lyrics app for MacOS and iOS. It searches for lyrics via free public APIs, then formats the output for line-by-line reading with configurable character limits. The code is on GitHub. There is no pre-compiled build, so you need a Mac with Xcode to build it from the repo.

How It Was Built

Ashley used GPT-4o Astra in Codex on a Pro plan, set to the extra high effort level. The full app took 3 prompts: one initial prompt did the heavy lifting, and two follow-up prompts refined the interface.

The initial prompt specified Swift 6 with strict concurrency, strict adherence to Apple’s API and human interface guidelines, a few keyboard shortcuts, and an interface overview. Everything else was left to the model: architecture, API selection, and all other decisions.

Why GPT Over Claude

Ashley makes a direct comparison worth noting. The vast majority of vibecoded projects you see use Claude. Ashley does not. The stated reasons:

  • Claude generates over-engineered code
  • Claude is too agreeable and does not check itself well
  • GPT follows instructions more precisely
  • GPT writes cleaner code with less baggage
  • GPT stays more focused in longer tasks and projects

The output is not described as perfect, but Ashley says it is better than expected. The takeaway from the experiment: a strong set of agent rules and a solid initial prompt can carry a project a long way, even without architectural guidance from the developer.

Stay on top of AI & Automation with BizStack Newsletter
BizStack  —  Entrepreneur’s Business Stack
Logo