Contents

1. Markdown / Writing presentations
2. Collaboration
3. Deploying / Keeping your data safe
4. Integrating LLMs
5. Evaluation / Tips

A few questions first

• Who here would consider themselves a developer ?
• Experience with code editors / IDEs ?
• More comfortable with editor than with other software ?
• Does text just feel faster/more efficient at this point?

  Markdown

(It's everywhere - you've most likely dealt with it at least once)

# Main title
## Heading 2 - You can make:

- *italic text*
- **bold text**
- ~~strikethrough~~
- lists
- [links](http://example.com)
- ![images](image_name_here)
- > Quotes

Why not write presentations as text ?

Introducing:  (Yuki Hattori)

• Markdown presentation ecosystem
• Comes as a command line tool or a VSCode editor extension
• Technically Just A Web Page $\implies$ can write HTML/CSS if you want to customize
• MarpX (Paolo Cunha) offers high quality ready-to-use themes
• If you know Markdown/HTML/CSS: No need to learn (yet another) tool!

Collaborating with others / keeping track of versions ?

"Well surely the editor can't do that ", I hear you say

• All the presentations are text-based versions of a file tree
• We need a tool to maintain versions of a text-based file tree...
• Ever heard of git?
• We already have a tool to collaborate on versions of text-based projects!

How do we make sure data is safe from big companies?

• There are online platforms like GitHub for this (but: this defeats the whole purpose)
• It's 2026, and servers are pretty cheap
• You can set up your own infrastructure in an afternoon
  • (less than 100 lines of docker-compose.yml)

One step further: hosting services on your own infrastructure

• I won't tell you how to set it up (plenty of guides online)
• TL;DR: On pushing changes $\rightarrow$ Auto-Render and Upload
• Here are some tools I'm using for my personal setup:

...isn't this really expensive, though?

• Not really!
• Domain name: ~10 EUR/year
• 24/7 VPS (200GB Storage, 6 vCPUs, 12GB RAM): 6.72 EUR/month
  • Also hosting other services on it and not even at capacity
  • I use Contabo, but Hetzner/OVH/Hostinger also work
• it's all yours, no need to worry about your data being accessed by third parties!
• Privacy does have a price, but so does exposing your personal data
  • Which one are you willing to pay?

...wasn't this supposed to integrate LLMs as well?

• It can be done as well!
• Head start when hosting models: Ollama (from Ollama Inc.)
• Automatically handles the download and execution of open models locally
• Works when handling content, since everything is text-based !
• We can integrate this into our IDE of choice and never have to worry about our data being sent to anyone

Pros and cons from experience

Tips

• (same as any AI augmented tool): Don't blindly trust the output, but use it as an assistant
• When using local models: try different models depending on your needs and hardware
• Don't be afraid to tinker around and see what works

Conclusion

• Presentation software / AI often feels "out of reach" and only hostable if you're a company and have the resources for it
• It doesn't have to be that way!
• Don't contribute to that narrative $\implies$ De-mystify the topic!
• Your data belongs to you; own it!

This presentation was done with MarpX too!

wget https://pages.ct.cozytren.ch/m1s2-eng-presentation/presentation.pdf
git clone https://gitea.ct.cozytren.ch/romane/m1s2-eng-presentation

GenAI use disclosure

(slide was appended after the presentation as per class instructions)

• qwen3-coder:30b, hosted locally using Ollama, has been used
• No use was made when generating the content
• The model has been used while proofreading the presentation for formatting and spelling issues but hasn't changed the content in any meaningful way