← All work
Case Study

sodigi·learn – AI wiki in 4 languages

7 courses · 90+ lessons · 4 languages

sodigi·learn – AI wiki in 4 languages

The challenge

A free learning platform on AI and vibecoding – in four languages at once, with no editorial team and no budget for one. Done the classic way, this volume of content and translations would have taken months. The task: build it all solo, with AI agents as the workforce and my own experience as the quality filter.

Approach & prompt process

  1. 1

    Structure before the first prompt: the map of 7 courses with lesson topics was fixed up front. The agent never got “write a course” – it got a specific lesson with a goal, an audience, and a tone.

  2. 2

    Every lesson ran through a draft → review → fix cycle: Claude Code generated an MDX draft, I checked facts and tone and returned notes as a list – until it reached a state worth publishing.

  3. 3

    Translations as a pipeline, not one giant prompt: lessons were translated one by one against a glossary, so tool names and terms sound the same across all four languages.

  4. 4

    Search via Pagefind: a static full-text index with no server. The agent built the integration; I verified the results with real queries.

  5. 5

    Raw versions of the lessons exist for AI agents – machine-readable page variants, so the content can be read by more than just humans.

Outcome

Claude CodeAstroMDXPagefind