Template 2: electric boogaloo

README.md

@@ -0,0 +1,48 @@
+# UL FRI DF journal template
+
+## How to use
+
+Everything that you need to edit is in the `edit_this` folder. You shouldn't have to touch anything else. The top of each file contains instructions for what to do.
+
+## Compiling
+
+The easiest way to compile this is using `tectonic`. Simply run `tectonic main.tex` in the root of the repository and you'll get a `main.pdf` file out.
+
+## Using LLMs
+
+I happened to be experimenting with LLMs a lot the week that I had to do this, so natually I wanted to see how useful they might be to automate some of this work.
+
+The first thing you'll probably want to do is textify all the pdfs so the LLM can read them.
+
+```sh
+find seminarske/*.pdf -exec pdftotext -raw {} \;
+```
+
+You'll also want to make a "razpored.txt" file with the works ordered by categories. If you got a pptx file, this will do:
+
+```sh
+libreoffice --headless --invisible --convert-to pdf razpored.pptx
+pdftotext razpored.pdf
+```
+
+Then you can have some fun with your robot intern. These prompts were tested in late 2025, in the Zed IDE, using the Claude Sommet 4 model. YMMV
+
+PDF includes:
+
+> The "seminarske" folder contains pdf journal submissions with their corresponding .txt version. The first few lines of each file contain the title. Write and execute a shell one-liner to extract the titles corresponding to the file names into your context.
+Read the "razpored.txt" file to map titles to categories.
+> Write the following latex command for each of the works, keeping their order from razpored.txt intact:
+> \includework
+> {Title of category}
+> {seminarske/name_of_the_file.pdf}
+> {Full title of the work}
+
+Summaries:
+
+> You now need to look through each of the files and extract the abstract and keywords. It should be in the first hundred lines.
+> For each work, output in the following form:
+>  \subsection{Full title of the work}
+>  Full, unchanged abstract goes here
+> \textbf{Keywords:} copy, the, keywords, here
+> For abstracts in non-english, use the same term they used for keywords (e.g. Ključne besede)
+> Process and output them in markdown blocks one at a time so you don't fill up your context window and make mistakes.