Tech

The growth website is live now.
Now future monthly learnings should be more complete.

I fixed cloudflare proxy that caused login issues (now wired locally), went live with the new website with new ansible scripts.

Started working on a library to edit xlsx files that does not break formatting (unlike openpyxl). I was surprised that there aren't any better libraries out there for python. Learnt the structure from python-docx and ppt from scanny, parts factory and using lxml to do surgical edits.

Excel is simlar to word document, all the contents are saved in xml and zipped. It also usese relationships to link all the files together.

I added support to pandas as well.

I will try to make it into a public library.

Continued working on my education website

Achievements:

• Design updates and improvements
• Added more content especially for the ODE/PDE course

Settled on Ubuntu as the production host for the Jonas sites — one server running the apps behind a reverse proxy.

Infra is described as Ansible playbooks so base setup, deploys, and rebuilds stay repeatable when migrating or standing up a new box.

Started a shared login service for the Jonas sites — one place to sign in, so edu, homepage, and later apps don’t each keep their own accounts.

Apps stay public where they should; private materials stay behind roles decided per app. No passwords stored on the client sites.

Perceptual Neutral:

hue 245, chroma 0.002. Tiny cool tint to counter the warm bias most monitors have at low brightness. Apple-HIG trick — looks like pure gray on your screen.

Clearly visualise and analyse the data first, think of a few ways that could solve it

Think what might affect valuation

Sharpe ratio -> volatility could be an important factor -> non linear parameters of it

Build a baseline model

Could use test data in training as well

Eye selection -> new adjustment layer -> Hue -> red channel reduce saturation -> add new adjustment layer -> Alt + Drag to have selection -> Hue -> colorize

Exercise notes for Large Scale AI.

Did Computer Vision Lectures

Engineering and optimizing large-scale AI systems on HPC/GPU clusters — model optimization, distributed workloads, profiling, and team projects.

Based on Racket like python but with contracts (controlling types and tests etc)

Did more lectures on Management of Digital Transformation

exiftool -time:all -a -G1 "/Users/jonas/Downloads/sth.pdf"

Lecture Notes on Management of Digital Transformation

ToDo:

• When there is no internet it should not update / but use the last message

• Daily plan strikethrough changes

Batch image expander for social media post

Dynamics HW 4

Paper Downloader

1. Scrape all the professor's names

2. Find their AuthorId

3. Find all their Papers

4. Todo: Download all their papers

Copy folder structure from Radix_Vorlag into Mandant server

The ultra sound will generate mini bubbles cleaning the items

Lower frequency generates bigger bubbles and cleans stronger. Higher frequencies are more gentle but cleans less well 40-60kHz is a good range

If there are airholes or fractures, it might damage the item

Do Not Clean:

• Emeralds, Opals, Pearls, Coral, Amber, Turquoise, Lapis lazuli, Malachite, Onyx, Tanzanite, Moonstone, Zircon

• Glued gems, Ultrasonics can loosen adhesives, Stones with fracture filling, oiling, or dyeing, Heat-sensitive gems

MacTeX for full 4Gb installation

I only took few notes on this course.

Goal:

• Automate Active Directory

• Automate Azure Permissions

• Automate Sharepoint

m = folium.map()

Need at least 7mil QR code. 50W 像素

Taobao 3/5mil, 100W pixel, 73-93mm DOF (depth of field), 380rmb

4 Stars Comet 4/7mil, 50W pixel, 300rmb

5 Stars Comet 3/4mil, 100W pixel, 500rmb

It is used for caching, Session Management, Real-time Analytics, Pub/Sub Messaging, Persistent Storage, Scalability

redis-cli to start

By default it is bind to localhost and also protectedmode (clients from other hosts can't connect), (read more about requirepass and ACL for user managment)

Redis should not be public. Better to use it as a storage service for a backend where users can connect

Redis uses binary TCP

DBeaver to visualize and edit your databases

Create multiple users for each application on MariaDB to have them separated, never use admin as it will have access to all databases

I can record log files in my.conf

general_log = 1
general_log_file = /var/log/mysql/mysql.log

But I need to make sure to have right file privilages set and that I use something like logroate to compress and move my files. As the logs contain crucial information and saved in plain text.

Download open source models through Ollama and run them locally

Deepseek R1 is 400gb the smaller ones might not be as good

Pretty easy to use, Deepseek has good documentation on how to use