Polishing the Markdown lingua-franca post: backing claims with papers and a new cover

A single-thread day spread across many parallel workspaces, all pulling on the same post.

tgmarinho-ai-website

Most of the day went into one bilingual post: "Markdown as the lingua franca of LLMs". The work was less about writing and more about defending the claims and tuning the voice.

I traced where the RAG advice came from (export to clean Markdown with tools like Docling, Marker, and MarkItDown, then chunk by headings and measure) and looked for real backing instead of Twitter folklore. I pulled in an arXiv paper and the LangChain Docling integration docs so the central points have a source, not just a vibe.

Then a tone pass. A "fine print" joke read badly, so I cut it. A line aimed at people who "want to go beyond Twitter guesswork" was too sarcastic, so I softened it. A subtitle was too long with tiny text, so I rewrote it. Every change applied to both the pt-BR and the English versions so the pair stays mirrored.

I also dug into JSON versus JSONL for LLMs: when each format actually wins, rather than assuming JSONL is always better. That nuance fed back into the post.

Finally, the cover. I settled on a contextual image for the article and wired it into the post frontmatter, then opened the PR.

A reminder that "finishing" a post is mostly the part after the draft: sourcing the claims and cutting the lines that try too hard.