Conference Talks → 🇵🇱

Accessing language models has meant either a chat window in a browser or the rather inconvenient approach of calling an API — with scripts, tokens, and handling JSON requests and responses. Fortunately, a third, and the only proper, way has finally arrived: CLI tools that work where I prefer to work — in the terminal. Claude Code, OpenAI Codex CLI, Gemini CLI and similar tools have become part of my everyday workflow. In this talk I will describe how I regularly use these interfaces, with particular focus on T_EX-related use cases: from generating and debugging macros, through supporting work with text, to automating various tasks — from very simple but often tedious ones to quite complex ones.

I will show concrete examples of use — both everyday and standard ones, as well as sometimes more wild experiments. I will also discuss what works well in this approach and what to watch out for. The presentation will include examples of integrating AI CLI interfaces not only with T_EX but also with other useful tools such as: git, make, python+venv, docker, html+js+css.

The talk will be practical in nature, illustrated with real-life examples. We will also attempt a live demonstration of some simple tasks, in direct interaction with the AI CLI tools being discussed.

Any typesetting system relies on two essential elements: fonts and the engine that arranges them into coherent and meaningful forms. In its early days, T_EX used bitmap fonts generated with METAFONT, which were never widely adopted outside the T_EX community. To reach a broader audience, these fonts had to be converted to commonly used formats. A major milestone was the 1992 release of Computer Modern in the PostScript Type 1 format by Blue Sky Research and Y&Y. This was a significant achievement, but it did not fully solve the problem, as emerging competing formats like TrueType and OpenType quickly appeared.

At the turn of the 20th and 21st centuries, the GUST e-Foundry team developed a font generation system that uses METAPOST as its main component. Over time, the system underwent substantial redesigns to keep pace with persistently evolving technologies and licensing requirements. The current version, named Fontplant, has reached a mature and reliable state. It generates all text fonts developed by GUST e-Foundry, i.e., Antykwa Półtawskiego, the Latin Modern family, and the T_EX Gyre collection. During the work on Fontplant, we were able to introduce some improvements to the fonts themselves.

Documentation is being finalized, with the goal of releasing Fontplant together with its METAPOST sources later this year. While no specific refinements are currently planned, experience shows that users may still propose ideas we will find hard to ignore.

Interslavic is a zonal auxiliary language created to facilitate communication among Slavic people. According to its creators — linguistics enthusiasts — sentences in Interslavic are intended to be understandable to nearly all Slavic speakers, even if they have never learned it before. I intend to discuss the history and basic features of this language, as well as the aspects necessary for typesetting texts in it using T_EX. Particular attention will be given to the hyphenation patterns I am developing.

It was a dark and chilly night, and I was reading my email. I had just received one that on first glance looked like a classic scam and I almost deleted it without reading it, but after I had closed my computer I suddenly realised that it mentioned LaT_EX. I went back to that email, read it for real this time, and there was no mistaking it: it was written by real lawyers who were actually looking to hire a T_EX specialist to help them in a court case that was ongoing. I replied to the email and started a conversation that led me to be hired as a T_EX expert in court, where I gave a testimony about T_EX engines and LaT_EX packages that was scrutinised by several lawyers and a judge, the latter then rendering a judgement where the use of T_EX played a central role. This is the story of T_EX’s day in court, my involvement, the questions I was asked about T_EX, and much more.

The T_EX

The Trial

The Judgement

Word division, or hyphenation, is a topic that has kept me busy for the better part of two decades, mostly at a T_EXnical level. I have however also had the opportunity to reflect and research the more typographic and linguistic aspects of this topic, and have come to realise that English contains good examples of what choices one has to make when choosing where to put breakpoints in a word.

This talk, entitled part two, is the third part of a multi-part series that I started in 2023 in BachoT_EX. During part one and a half in 2024, the audience was invited to consider the following hyphenated words and attempt to explain them:

• crit-i-cism
• an-eu-rysm
• lig-a-ture
• sha-king
• as-tron-o-my
• as-tro-nom-i-cal
• dem-o-crat
• de-mo-cra-cy
• dem-o-crat-ic
• phe-nom-e-nol-o-gy
• e-nu-mer-ate
• pi-rate
• riv-ers
• ri-vals
• bi-son
• pris-on
• knowl-edge

• proc-ess
• proj-ect
• pres-ent

• pro-cess
• pro-ject
• pre-sent

Some of the breakpoints above may look surprising, but they are nonetheless correct according to mainstream American typographic tradition: there is a simple rule that explains them all. Everything will be revealed during my talk.

Eighteen years ago, Mojca Miklavec and myself had a conversation in BachoT_EX that sparked an effort to rationalise and improve the way hyphenation patterns were handled in T_EX Live. This talk will give a summary of that journey, and our plans for the future.

This paper presents a T_EX-first approach to managing address data and mail merge workflows using BibLaT_EX. The proposed solution, AddressBookLite, treats a .bib file as a semantic data model rather than a simple list of references.

The core of the approach is an extensible data model (.dbx) combined with a custom formatting driver (.bbx) that produces only meaningful output lines, eliminating empty or inconsistent elements. The system architecture consists of three layers: data (.bib), formatting logic (biblatex), and a lightweight local application written in Python (Flask) for editing and validating records.

The paper also discusses the role of AI as a technical assistant in the design process, particularly in data model design, detection of semantic inconsistencies, and iterative debugging of formatting logic.

The proposed approach demonstrates that T_EX remains a powerful environment for structured data processing, and that combining T_EX, AI, and domain expertise enables flexible, portable, and sustainable solutions.

Accessibility on the web has improved considerable over the last years, with screen readers being able to read out pages without great mishaps and problems. But when it comes to mathematics and more general STEM, the situation has only very recently started to change with the introduction of MathML and its support in browsers.

In this talk we present the basics of MathML, open source tools to create it, and look at open source fonts that can be used to render MathML on the web.

arXiv is the world's largest and oldest scientific preprint server, and a champion of open science. Started in 1991, arXiv presently holds more than 2.4 million articles and is growing at an ever-increasing rate.

With over 30 years of history, arXiv contains T_EX documents that span most of the lifetime of T_EX itself. Through today, we try to be able to recompile all documents. For that, we keep multiple versions of T_EX distributions, starting from teT_EX 2 up to, currently, T_EX Live 2023.

In this talk we will report about the problems we are facing with keeping all these versions up and running, deal with the ever-increasing set of packages that get obsoleted, and all the other niceties of having to support very old software.

We will also report about our current pipelines, which tools we have settled for watermarking, pdf concatenation, and other operations we need.

For more than fifty years I have worked with computers, including four decades with T_EX-related systems. During this time, I have accumulated large digital archives. Once they reached a certain scale, their organization and analysis became difficult without dedicated tools.

In the talk I will present the story of designing and building a tool for “digital archaeology”. I implemented it in close collaboration with ChatGPT.

I will discuss the motivations behind the project, the development process, and the role of AI, among other things, in accelerating the implementation. I will also describe how, during the work, both sides made some mistakes and how they were gradually identified. The final result proved fully successful and is now used in practice.

Part 2: Implementation

The program xdvipsk extends dvips by adding support for:
(1) more flexible inclusion of bitmap images,
(2) new specials with prefixes mapline and mapfile,
(3) OpenType fonts,
(4) ToUnicode CMaps for Type 1 PostScript and OpenType fonts,
(5) Lua callbacks.

The results of our long development were recently recognized by inclusion of xdvipsk in T_EX Live 2026, so we are going to present these extensions. Our earlier paper (TUGboat 38:2 197–201) described extensions (1) and (3), this article describes changes made from that time.

The changes to xdvipsk were necessary to produce accessible publications in workflow TEX → DVI → PS → PDF. Our solution to accessibility problem required much more changes in LuaT_EX kernel and LaT_EX classes, but they are out of scope of this publication.

The minimum required for xdvipsk to function with extensions (1), (2) and (4) is to do the first workflow's step T_EX → DVI by LuaT_EX using the bundle of LaT_EX package xdvipsk-support. Then the step xdvipsk: DVI → PS provides Unicode values for all document's characters, that were included in `glyphs to unicode' maps. Some maps come from fonts (OpenType), some are built in the xdvipsk code, others must be provided externally. The final PDF result depends also on the program used in the final step PS → PDF.

Kilkadziesiąt okładek książek i czasopism tworzonych przez Andrzeja Tomaszewskiego. Przykłady odzwierciedlają nie tylko założenia projektowe autora, ale będzie też mowa o uwarunkowaniach poligraficznych i introligatorskich oraz o wpływie gustów wydawców na ostateczną formę. Zbiór nie ma charakteru rozprawy o okładkach jako formach opakowania publikacji. Jest tylko autorskim wyborem projektów powstających na przestrzeni 30 lat.

Na koniec próba tworzenia okładkowej grafiki w ChatGPT.