Why LaTeX

Why do you use LaTeX?

I started with LaTeX because we were building Jobward in elixir, a tool for generating targeted résumés to get through the ATS gatekeepers. We needed to represent typography created by our designer in a way I could actually trust to a document template without making the user edit the page manually. Word and the HTML/CSS résumé builders gave me documents that looked almost right, until a recruiter opened them on a different machine and the layout fell apart. Converting HTML pageless, dynamically formatted content into PDF is challenging. But, LaTeX gave me a typeset résumé that was pixel-perfect everywhere and a source file I could diff in git. This is a great example of how LaTeX can be used to create a tool that is not just a document, but a tool for creating documents.

Once I saw what it could do for pages of a résumé, I wanted to know what else it could do. To practice I started digitizing out-of-print books in Buckminster Fuller's library into LaTeX manuscripts. That pushed me to learn typesetting math equations, chapters, indexes, glossaries, figures, tables, and references. Now I have a toolchain building all the eBook formats for my eReaders. Now I have over a hundred books, all compiled from .tex source into PDF, Markdown, EPUB, and Kindle.

What do you love about it?

That it's a compiler, not a word processor. My manuscripts are plain text, version-controlled, and every output format — print PDF, EPUB3, Kindle KPF, even the markdown that becomes a wiki page — comes from the same source. scrbook already knows what a frontmatter, mainmatter, and backmatter are; that glossaries and biblatex handle the bookkeeping; that I can write one sentence per line and let latexindent and Prettier worry about wrapping. And I love that when a Fuller book has a tetrahedron diagram in a figure, TikZ can redraw it cleanly instead of me scanning a smudged 1975 print into a dithered image.

What frustrates you?

Package management. Keeping a TeX environment identical across machines — laptop, CI runner, a collaborator's Mac, a Docker container — is a mess. tlmgr update --all on one box and a manuscript that compiled yesterday fails today because a package moved or a dependency shifted. There's no package.json lockfile moment of clarity; you're juggling MacTeX vs. BasicTeX, frozen TeX Live years, and packages installed ad hoc until something finally works. Error messages don't help much either — a missing brace inside a minted block ten files deep in \input{} calls becomes a 200-line log that points to line 1 of manuscript.tex. And the docs are scattered across CTAN PDFs, TUGboat articles, and Stack Exchange answers from 2009 that reference deprecated commands. Every new project starts with thirty minutes of "is this the modern way to do X, or the 1998 way?"

Favorite package?

biblatex with biber — once I learned it, citation management stopped being a chore and became a feature. Close runners-up: glossaries (which makes Fuller's invented vocabulary — ephemeralization, synergetics, tensegrity — feel like first-class objects in the book), minted for code listings, and KOMA-Script's scrbook for being the sane default book class should have been.

TikZ or another drawing tool?

TikZ, almost always. For Fuller's geometry — vector equilibria, jitterbug transformations, geodesic frequencies — TikZ with shapes.geometric and a little tikz-3dplot produces diagrams that scale perfectly into both print PDF and EPUB. When something is too organic for TikZ I'll draw it in another tool and import as SVG/PDF, but I default to TikZ because the diagram lives in the source and survives every re-typeset.

Best editor?

JetBrains WebStorm with the TeXiFy IDEA plugin. I came to it from web development and never left. Soft-wrap on .tex, latexindent + Prettier (with the LaTeX plugin) on save, git baked in, and inspections tuned so "one sentence per line" doesn't get flagged. PragmataPro as the monofont. TeXShop and VS Code both have their charms, but TeXiFy's structure view and forward/inverse search to Skim is what made long manuscripts manageable.

Biggest LaTeX success or disaster?

Success: the whole Buckyverse pipeline. A book starts as a physical object on a CZUR ET24 Pro scanner, becomes DOCX, then pandoc converts it to .tex, then I clean up scannos and add macros, glossary entries, and figures, and the same source compiles out to PDF for print, EPUB3 via tex4ebook, KPF via Kindle Previewer, and Markdown for the wiki. Synergetics, Critical Path, Operating Manual for Spaceship Earth — books that were out of print or locked in image-scan wrapped pdfs now have living, searchable, hyperlinked editions because LaTeX gave me one source of truth.

Disaster: the day I learned what happens when you change a font family deep in a 600-page manuscript and the \index page numbers shift by one across the entire backmatter. Or the day a \textemdash mid-bibliography blew up biber in a way that took an afternoon to trace. Every LaTeX disaster, in retrospect, is a feature: the system told me the truth about my document; I just didn't want to hear it yet.

Anything else related to LaTeX?

LaTeX taught me a deeper lesson than typography. Fuller talks about ephemeralization — doing more and more with less and less until eventually you can do everything with nothing. Plain .tex files in git, compiled into every format a reader might want, is ephemeralization for books. I picked up LaTeX to make a résumé look right; I'm still using it because it turned out to be the right tool for preserving a library.

The next experiment is Codexis. Today the pipeline is scan → TIFF → OCR → DOCX → pandoc → .tex → proofread forever. Pure OCR gives you text, but it flattens structure — it doesn't know a displayed equation from a footnote from a figure caption, and every page leaves you hunting scannos (rn for m, l for 1, em dashes turned into hyphens). I'm testing whether a local vision LLM — Qwen 3 VL, run on my own hardware via Ollama — can skip the middle formats and read a page .tiff directly, emitting LaTeX that already respects the layout: \chapter{}, \begin{equation}, \footnote{}, \includegraphics{}. The model sees the page the way a human proofreader does, not as a bag of characters. Early results suggest it cuts the manual cleanup substantially compared to OCR-then-pandoc, especially on math-heavy Fuller pages where Tesseract never had a chance. It's still an experiment — you still validate every page — but the goal is the same as always: one faithful .tex source, every format downstream, less time fixing what the machine misread.