From React to Qt Widgets Using AI (cristianadam.eu)

The author ported the llama.vim functionality into a Qt Creator plugin (llama.qtcreator v2.0.0), using AI assistants (gpt-oss 20B and Qwen3) to help translate TypeScript/React web UI concepts into C++/Qt Widgets. The goal expanded from code completion to a chat interface that renders model-generated markdown via md4c and litehtml. This project demonstrates practical, AI-accelerated code migration: models can produce helpful skeletons and targeted fixes but struggle with large files, make subtle bugs, and need very concrete prompts to be effective. Key technical takeaways: the original web UI used IndexedDB via Dexie, so storage was mapped to Qt’s built-in SQLite—an easy schema fit but a bug surfaced from naive string-based SQL generated by the model. Rendering required adapting Qt Creator’s QLiteHtmlWidget (a litehtml wrapper) but its QAbstractScrollArea inheritance prevented incremental message growth; using internal DocumentContainer APIs allowed rendering but lacked selection and incremental updates, so the author reverted to QLabel and later to QTextBrowser to implement incremental section rendering and conversation search. UI polish included converting SVG icons to a TrueType font to fix resolution and dark-theme issues. The write-up is a pragmatic guide to where LLMs help in code conversion and where human expertise and architecture tweaks remain essential.