Python

𝐎𝐧𝐥𝐲 𝟐% of my 𝐩𝐫𝐨𝐦𝐩𝐭𝐬 contain 𝐩𝐫𝐨𝐟𝐚𝐧𝐢𝐭𝐲

Written by marcelpetrick on August 11, 2026

Testing Muse Glimmer lead to some frustration, so I started the next prompt with something I wouldn’t write professionally.
And this led to the question: how often do I do this? I would have guessed that 10% of my machine-human interactions contain some swear words.

But why guess when you can evaluate the data and produce hard facts?

Therefore I prepared 🫙 𝐀𝐠𝐞𝐧𝐭𝐢𝐜 𝐒𝐰𝐞𝐚𝐫 𝐉𝐚𝐫 as project on GitHub, which evaluates both your 𝐂𝐨𝐝𝐞𝐱 and 𝐂𝐥𝐚𝐮𝐝𝐞 𝐂𝐨𝐝𝐞 logs and gives you on the CLI a short two liner. And additionally you receive an HTML page in nice green IBM 3279-style. All locally, no AI or models involved, just some Python and HTML.
It will reveal to you how many swearwords were used, their frequency, streaks without swearing, which hour of the day the tension rises high, etc ..

What do you think? Anyone has a higher score to offer? Leave me a comment.

GitHub: https://github.com/marcelpetrick/AgenticSwearJar

PDF with full feature view:

agentic-swear-jar-linkedin

Screenshot:

𝐂𝐥𝐨𝐮𝐝 𝐀𝐏𝐈𝐬 𝐚𝐫𝐞 𝐞𝐱𝐩𝐞𝐧𝐬𝐢𝐯𝐞. 𝐖𝐚𝐭𝐞𝐫𝐦𝐞𝐥𝐨𝐧 𝐢𝐬 𝐧𝐨𝐭. 🍉

Written by marcelpetrick on July 15, 2026

Vision models grow in their capabilities while the execution demands way fewer resources.

Means: my sales slip scanner-project from two years ago needed an upgrade. Before I was using GPT-4 Vision via the @OpenAI API, which costs you control over your data and a few cents for every analysis.

Now, after intensive benchmarking across a range of 15 models (with 𝐎𝐥𝐥𝐚𝐦𝐚) and some surprising regressions (newer harness doesn’t mean the models execute faster; it can also mean they suddenly run into model load errors…), I put my money on 𝐐𝐰𝐞𝐧𝟑.𝟓:𝟒𝐁. A casual language model with a 𝐯𝐢𝐬𝐢𝐨𝐧 𝐞𝐧𝐜𝐨𝐝𝐞𝐫. My own benchmark over the selected models against the huge sample size of three receipts: all three were successfully evaluated. 😉

Then I built a 𝐡𝐨𝐭-𝐟𝐨𝐥𝐝𝐞𝐫-𝐜𝐨𝐧𝐜𝐞𝐩𝐭 around the script, and it works (thank you @Mr Zahorsky — without whom I would have never come across that idea). For instance: in less than ten seconds, 13 samples were processed, and you get the result sum plus a fancy HTML overview for comparison.

𝐃𝐚𝐭𝐚 𝐞𝐧𝐭𝐫𝐲 can be done for cheap: give the kids the smartphone, snap all receipts, throw out some watermelon slices 🍉😉

If you want to run the tool as well, check my GitHub: https://github.com/marcelpetrick/sales-slip-scanner-ng. As said: a local GPU is the only thing you need; 𝐧𝐨 𝐜𝐥𝐨𝐮𝐝, 𝐧𝐨 𝐀𝐏𝐈 𝐤𝐞𝐲𝐬, 𝐧𝐨 𝐜𝐫𝐞𝐝𝐢𝐭 𝐜𝐚𝐫𝐝. Everything is now self-hosted. You’ll also find the benchmark of the vision models there, including the results.

𝐓𝐡𝐞 𝐛𝐢𝐠 𝐪𝐮𝐞𝐬𝐭𝐢𝐨𝐧 𝐢𝐬: 𝐰𝐡𝐚𝐭 𝐬𝐡𝐨𝐮𝐥𝐝 𝐈 𝐚𝐮𝐭𝐨𝐦𝐚𝐭𝐞 𝐧𝐞𝐱𝐭 𝐰𝐢𝐭𝐡 𝐚 𝐥𝐨𝐜𝐚𝐥 𝐯𝐢𝐬𝐢𝐨𝐧 𝐦𝐨𝐝𝐞𝐥? 𝐖𝐡𝐚𝐭 𝐰𝐨𝐮𝐥𝐝 𝐲𝐨𝐮 𝐝𝐨?

#neverstoplearning

𝐌𝐲 𝐚𝐩𝐩𝐬 𝐚𝐫𝐞 𝐧𝐨𝐭 𝐚𝐛𝐚𝐧𝐝𝐨𝐧𝐰𝐚𝐫𝐞!

Written by marcelpetrick on July 10, 2026

Yesterday, #RitterRadar got some updates: first, all dependencies were refreshed (no more CVEs); then I fixed some bugs I had noticed over the past few weeks; and finally, the best news: a new feature that gives you a copyable list of the events matching your current filter.

So, for instance, if you live near Potsdam and want to see all medieval events within 50 km over the next two months – you get this:

——-

——-

#FastAPI #WebScraping #OpenStreetMap #Geocoding #OpenSource #EventDiscovery

URL to the GitHub-project: https://github.com/marcelpetrick/RitterRadar if you haven’t given it a try yet ⚔️🐉

⚔️🐉 𝗥𝗶𝘁𝘁𝗲𝗿𝗥𝗮𝗱𝗮𝗿 🛡️🏰🏇

Written by marcelpetrick on June 27, 2026

𝘏𝘦𝘢𝘳𝘬𝘦𝘯, 𝘨𝘰𝘰𝘥 𝘵𝘳𝘢𝘷𝘦𝘭𝘭𝘦𝘳, 𝘢𝘯𝘥 𝘭𝘦𝘯𝘥 𝘵𝘩𝘪𝘯𝘦 𝘦𝘢𝘳!

𝘙𝘪𝘵𝘵𝘦𝘳𝘙𝘢𝘥𝘢𝘳 𝘪𝘴 𝘢 𝘤𝘶𝘯𝘯𝘪𝘯𝘨 𝘪𝘯𝘴𝘵𝘳𝘶𝘮𝘦𝘯𝘵, 𝘧𝘰𝘳𝘨𝘦𝘥 𝘪𝘯 𝘵𝘩𝘦 𝘧𝘪𝘳𝘦𝘴 𝘰𝘧 𝘗𝘺𝘵𝘩𝘰𝘯, 𝘵𝘰 𝘢𝘪𝘥 𝘵𝘩𝘦𝘦 𝘪𝘯 𝘵𝘩𝘺 𝘯𝘰𝘣𝘭𝘦 𝘲𝘶𝘦𝘴𝘵: 𝘵𝘩𝘦 𝘥𝘪𝘴𝘤𝘰𝘷𝘦𝘳𝘺 𝘰𝘧 𝘮𝘦𝘥𝘪𝘦𝘷𝘢𝘭 𝘮𝘢𝘳𝘬𝘦𝘵𝘴, 𝘙𝘦𝘯𝘢𝘪𝘴𝘴𝘢𝘯𝘤𝘦 𝘧𝘢𝘪𝘳𝘴, 𝘝𝘪𝘬𝘪𝘯𝘨 𝘴𝘱𝘦𝘤𝘵𝘢𝘤𝘭𝘦𝘴, 𝘢𝘯𝘥 𝘊𝘩𝘳𝘪𝘴𝘵𝘮𝘢𝘴 𝘳𝘦𝘷𝘦𝘭𝘳𝘪𝘦𝘴 𝘰𝘧 𝘵𝘩𝘦 𝘢𝘯𝘤𝘪𝘦𝘯𝘵 𝘴𝘵𝘺𝘭𝘦 – 𝘢𝘤𝘳𝘰𝘴𝘴 𝘵𝘩𝘦 𝘎𝘦𝘳𝘮𝘢𝘯 𝘭𝘢𝘯𝘥𝘴 𝘢𝘯𝘥 𝘣𝘦𝘺𝘰𝘯𝘥.

𝘛𝘩𝘪𝘴 𝘵𝘰𝘰𝘭 𝘥𝘰𝘵𝘩 𝘥𝘪𝘴𝘱𝘢𝘵𝘤𝘩 𝘵𝘪𝘳𝘦𝘭𝘦𝘴𝘴 𝘸𝘦𝘣-𝘤𝘳𝘢𝘸𝘭𝘪𝘯𝘨 𝘢𝘨𝘦𝘯𝘵𝘴 𝘪𝘯𝘵𝘰 𝘵𝘩𝘦 𝘷𝘢𝘴𝘵 𝘥𝘪𝘨𝘪𝘵𝘢𝘭 𝘸𝘪𝘭𝘥𝘦𝘳𝘯𝘦𝘴𝘴. 𝘛𝘩𝘦𝘺 𝘳𝘦𝘵𝘶𝘳𝘯 𝘭𝘢𝘥𝘦𝘯 𝘸𝘪𝘵𝘩 𝘵𝘪𝘥𝘪𝘯𝘨𝘴 𝘰𝘧 𝘧𝘰𝘳𝘵𝘩𝘤𝘰𝘮𝘪𝘯𝘨 𝘦𝘷𝘦𝘯𝘵𝘴 – 𝘵𝘩𝘦𝘪𝘳 𝘯𝘢𝘮𝘦𝘴, 𝘵𝘩𝘦𝘪𝘳 𝘥𝘢𝘵𝘦𝘴, 𝘵𝘩𝘦𝘪𝘳 𝘸𝘩𝘦𝘳𝘦𝘢𝘣𝘰𝘶𝘵𝘴 — 𝘢𝘯𝘥 𝘴𝘵𝘰𝘳𝘦 𝘢𝘭𝘭 𝘸𝘪𝘵𝘩𝘪𝘯 𝘢 𝘭𝘰𝘤𝘢𝘭 𝘵𝘳𝘦𝘢𝘴𝘶𝘳𝘺 𝘰𝘧 𝘚𝘘𝘓𝘪𝘵𝘦.

𝘜𝘱𝘰𝘯 𝘵𝘩𝘪𝘯𝘦 𝘰𝘸𝘯 𝘮𝘢𝘤𝘩𝘪𝘯𝘦 𝘪𝘵 𝘥𝘰𝘵𝘩 𝘳𝘦𝘯𝘥𝘦𝘳 𝘢 𝘮𝘰𝘴𝘵 𝘣𝘦𝘢𝘶𝘵𝘪𝘧𝘶𝘭 𝘪𝘯𝘵𝘦𝘳𝘢𝘤𝘵𝘪𝘷𝘦 𝘮𝘢𝘱, 𝘸𝘩𝘦𝘳𝘦𝘶𝘱𝘰𝘯 𝘵𝘩𝘰𝘶 𝘮𝘢𝘺𝘦𝘴𝘵 𝗳𝗶𝗹𝘁𝗲𝗿 𝗯𝘆 𝗽𝗲𝗿𝗶𝗼𝗱 (𝘸𝘩𝘪𝘤𝘩 𝘮𝘰𝘯𝘵𝘩𝘴 𝘵𝘩𝘰𝘶 𝘸𝘪𝘴𝘩𝘦𝘴𝘵 𝘵𝘰 𝘴𝘶𝘳𝘷𝘦𝘺) 𝘢𝘯𝘥 𝗳𝗶𝗹𝘁𝗲𝗿 𝗯𝘆 𝘀𝗽𝗮𝗰𝗲 (𝘵𝘩𝘺 𝘩𝘰𝘮𝘦 𝘱𝘰𝘴𝘪𝘵𝘪𝘰𝘯 𝘢𝘯𝘥 𝘢 𝘳𝘢𝘥𝘪𝘶𝘴 𝘰𝘧 𝘵𝘩𝘺 𝘤𝘩𝘰𝘰𝘴𝘪𝘯𝘨, 𝘧𝘳𝘰𝘮 𝘢 𝘴𝘵𝘰𝘯𝘦’𝘴 𝘵𝘩𝘳𝘰𝘸 𝘵𝘰 1024 𝘭𝘦𝘢𝘨𝘶𝘦𝘴). 𝘊𝘭𝘪𝘤𝘬 𝘶𝘱𝘰𝘯 𝘢𝘯𝘺 𝘮𝘢𝘳𝘬𝘦𝘳 𝘵𝘰 𝘭𝘦𝘢𝘳𝘯 𝘵𝘩𝘦 𝘧𝘶𝘭𝘭 𝘱𝘢𝘳𝘵𝘪𝘤𝘶𝘭𝘢𝘳𝘴. 𝘛𝘩𝘦 𝘤𝘳𝘢𝘸𝘭𝘦𝘳 𝘳𝘶𝘯𝘯𝘦𝘵𝘩 𝘪𝘯 𝘵𝘩𝘦 𝘣𝘢𝘤𝘬𝘨𝘳𝘰𝘶𝘯𝘥; 𝘵𝘩𝘦 𝘮𝘢𝘱 𝘳𝘦𝘧𝘳𝘦𝘴𝘩𝘦𝘵𝘩 𝘰𝘯 𝘪𝘵𝘴 𝘰𝘸𝘯 𝘢𝘤𝘤𝘰𝘳𝘥.

𝘕𝘰 𝘤𝘭𝘰𝘶𝘥. 𝘕𝘰 𝘢𝘤𝘤𝘰𝘶𝘯𝘵𝘴. 𝘕𝘰 𝘥𝘢𝘵𝘢 𝘭𝘦𝘢𝘷𝘦𝘵𝘩 𝘵𝘩𝘺 𝘮𝘢𝘤𝘩𝘪𝘯𝘦. 𝘛𝘩𝘶𝘴: 𝘢 𝘩𝘦𝘭𝘱𝘦𝘳 𝘵𝘰𝘰𝘭 𝘵𝘩𝘢𝘵 𝘤𝘳𝘢𝘸𝘭𝘦𝘵𝘩 𝘵𝘩𝘦 𝘸𝘦𝘣 𝘸𝘪𝘵𝘩 𝘤𝘶𝘴𝘵𝘰𝘮 𝘤𝘳𝘢𝘸𝘭𝘦𝘳𝘴, 𝘱𝘳𝘦𝘴𝘦𝘯𝘵𝘦𝘵𝘩 𝘵𝘩𝘦 𝘧𝘪𝘯𝘥𝘪𝘯𝘨𝘴, 𝘢𝘯𝘥 𝘢𝘭𝘭𝘰𝘸𝘦𝘵𝘩 𝘵𝘩𝘦𝘦 𝘵𝘰 𝘴𝘪𝘧𝘵 𝘵𝘩𝘦𝘮 𝘣𝘺 𝘵𝘪𝘮𝘦 𝘢𝘯𝘥 𝘣𝘺 𝘥𝘪𝘴𝘵𝘢𝘯𝘤𝘦. 𝘕𝘰𝘵𝘩𝘪𝘯𝘨 𝘮𝘰𝘳𝘦, 𝘯𝘰𝘵𝘩𝘪𝘯𝘨 𝘭𝘦𝘴𝘴 – 𝘣𝘶𝘵 𝘧𝘢𝘴𝘩𝘪𝘰𝘯𝘦𝘥 𝘸𝘪𝘵𝘩 𝘤𝘢𝘳𝘦.

——-

Also: Wir besuchen recht gerne Mittelaltermärkte. Jetzt ist es so, dass die Schaustellergruppen umherziehen beziehungsweise bestimmte Events an historischen Orten stattfinden. Ich wohne nicht an einem historischen Ort.

Also sucht man eine der vielen Seiten im Netz auf, die Termine auflisten. Diese sind über die Jahre gewachsen, und viele Seiten bieten auch Komfortfunktionen wie „Filter nach Datum oder Bundesland“ an, aber:
* 𝗙𝗿𝗮𝗴𝗺𝗲𝗻𝘁𝗶𝗲𝗿𝘂𝗻𝗴: Nicht alle Seiten zeigen „alle Events“
* Ein Filter nach Bundesland hat das Problem, dass man alles einzeln sichten muss, wenn man auch gewillt ist, Landesgrenzen zu überschreiten (oder eben ungefiltert, yada yada)
* 𝗙𝗶𝗹𝘁𝗲𝗿𝘂𝗻𝗴 𝗻𝗮𝗰𝗵 𝗭𝗲𝗶𝘁𝗿𝗮𝘂𝗺 (Vergangenes interessiert mich nicht)

Also, wie man sieht: ein komplexer Task, wenn man sich einen Plan für die nächsten Monate machen möchte. Durchaus lösbar, aber wie Clarence Bleicher sagt: „[..] 𝘁𝗵𝗮𝘁 𝘄𝗵𝗲𝗻 𝗵𝗲 𝗵𝗮𝗱 𝗮 𝗵𝗮𝗿𝗱 𝗷𝗼𝗯, 𝗵𝗲 𝘄𝗼𝘂𝗹𝗱 𝗮𝘀𝘀𝗶𝗴𝗻 𝗶𝘁 𝘁𝗼 𝗮 ‘𝗹𝗮𝘇𝘆 𝗺𝗮𝗻’ 𝗯𝗲𝗰𝗮𝘂𝘀𝗲 𝘁𝗵𝗲𝘆 𝘄𝗼𝘂𝗹𝗱 𝗳𝗶𝗻𝗱 𝗮𝗻 𝗲𝗮𝘀𝘆 𝘄𝗮𝘆 𝘁𝗼 𝗱𝗼 𝗶𝘁.“ – kein Job für mich 🙈

Nun bin ich auch KEIN Webentwickler. Aber das sollte mich nicht daran hindern, „Ritter Radar“ zu entwickeln.

Eine App, die aus mehreren Webcrawlern sowie Front- und Backend besteht. Alle bekannten Mittelaltermarkt-Seiten werden durchsucht, die Ergebnisse kommen in eine lokale Datenbank, und dann hat man die Möglichkeit, in einem Web-UI nach Entfernung zum Startort und Zeitraum zu filtern. Easy peasy.

BeautifulSoup, OpenStreetMap, Python, SQLite, Uvicorn, ein bisschen HTML und CSS … und fertig.

Ich habe es bisher nicht auf ein Web-Deployment getrimmt, weil ich die Arbeit der Ersteller der Mittelaltermarkt-Seiten honoriere. Wenn ich jetzt so ein Portal hosten würde, wäre ich wie Google mit den KI-Suchergebnissen: Eine Zusammenfassung drängt sich zwischen Nutzer und Ersteller und fängt so Webseiten-Traffic ab.

Daher: Einfach auf https://github.com/marcelpetrick/RitterRadar das Repository klonen und selbst bootstrappen: zwei Shell-Skripte ausführen, fertig ist der Lack. ⚔️🐉

grip: render Markdown as PDF

Written by marcelpetrick on May 6, 2021

.. and other things, where you assumed it should be quite easy. ..

Wrote a short guide how to verify some information in Markdown. Local rendering works (most of the time via PyCharm or online at Github).
Now: how export it as PDF, because I realized that the receiver might not be able to display it properly.

* printing from PyCharm: failed
* VisualStudio-Plugin: no VS, no plugin
* any of the *nix-ways: not possible at that moment
* using a web-renderer: not allowed, because confidental data

UFF!

Python to the rescue!

pip install grip  
grip file.md

1 2	pip install grip grip file.md

Grip prepares a local flask server, where you receive a localhost:<randomport> url and just open it with the browser of your choice and then print as PDF.

Project Euler – mathematical riddles which require some programming skills

Written by marcelpetrick on January 14, 2021

Took me a while to write about this, but I really love Project Euler. The page is a collection of math challenges, which require some programming (I saw just one which could have been computed by a closed formula without any help). The first ten are quite easy to solve and are more commonly known math problems. Prime numbers and combinatorics play a strong role. But then the difficulty rises quite quickly. Usually it takes me one to two hours to write a Python solution for one. If I would – like I should – write unit-tests for each single method and not for a few selected one, then I guess 50% more.

It’s great: each problem is a closed, separate problem, which requires some algorithmic thinking and – of course – some proper implementation. If you chose the wrong path time or space complexity will kill your ambitions quite quickly. But proper solutions are computed most of the time in less than a minute.
Most of the time I rely on basic python structures and common libraries. But I’ve also given NumPy, itertools, etc. a try. Speeds up the process quite a bit.
My next goal is to fix problem 47, because then I’ve handed in solutions for all of the first fifty problems.
The highest challenge (with also the highest difficulty level (for me) so far) was problem 668. Due to a really big (80 GiByte!) boolean array the computer had a hard time swapping memory. So it took almost 36 hours to finish. By the way: less than 900 people worldwide have solved this issue #tinyflakeofpride

Of course, several geniuses have dedicated pages to optimal solution strategies. Which is a nice idea. But I avoid them. Most of the times stepping back, thinking without a display about the problem and if the chosen approach was a good one, is more helpful. A solution by cheating is nothing which renders any reward.

https://github.com/luckytoilet/projecteuler-solutions

https://euler.stephan-brumme.com/24/

micro:bit v2 arrived :)

Written by marcelpetrick on January 4, 2021

Today (finally) my micro:bit v2 arrived. Had to unwrap it immediately after dinner and play around with the speech synthesis👌🏻 Some lines of microPython and the things got heated.
If you’re not creating anything nowadays, then it’s your own fault 🐒

simple webscraper for last.fm with BeautifulSoup

Written by marcelpetrick on September 29, 2020

tl;dr:
Simple webscraper with Python and BeautifulSoup for one user’s favorite tracks (‘loved songs’) at last.fm.
Repository: github

full text:
Looks like last.fm is shutting down its services (one feature at at time, lol). They started this process more or less ten years ago.
I’ve realized that I would miss my curated list of favorite tracks and I am also very bad at remembering, so … let’s automate the process of grabbing that information from their public page. I know, they offer a REST API, but I wanted to use once BS4.
Since I had somehow two free hours fourteen days ago, I went full-speed to some tutorials, played with the get-requests and how to parse. And then spent the last minutes parsing the received “artist+track”-string into something usable. Alltogether four hours were spent and I am amazed by the result. Of course, by leveraging three quite powerful libraries (beautifulsoup4, requests, lxml) and skipping TDD (;) I’ve reached the goal quite fast. And since the script works (my 1500 loved songs are scraped in less than 60 seconds), I will also not spend additional effort to make it “pretty”.

“[Full Day Workshop] Kubeflow + BERT + GPU + TensorFlow + Keras + SageMaker”

Written by marcelpetrick on September 27, 2020

I’ve just spent the last eight hours attending a workshop about #SageMaker, #AutoPilot, #BERT, #Athena, #TensorFlow, #Spark, [..] and I am feeling a bit light-headed.
Of course, the talk and guidance given by @AntjeBarth and @ChrisFregly was really well prepared, but if you’re just a ML-beginner (like me) and if then over 9000 of new technologies drop, you have to work hard to follow the fast paced event.
Of course, I started my ML-journey in the summer of 2019, but it was more focussed on image-processing, not #NLP. I worked before with #Python, #Jupyter notebooks, TensorFlow and #Keras, but that whole SageMaker-thing was new to me.
And I see the potential: instead of running the stuff locally, you prepare, prototype and run your ML-app inside Amazon’s infrastructure. And that AutoPilot, which helps to quickstart the prototyping by trying several preprocessing-steps and models for you on your data, looks promising. Will definitely give it a second look.
Notes can be found at: https://github.com/marcelpetrick/KubeFlow_BERT_GPU_TensorFlow_Keras_SageMaker_Workshop (need lots of polishing, as always)

Crazy times we live in! And I am thankful for this block of time on a weekend 🙏

PyQt: GroundSpace

Written by marcelpetrick on September 4, 2020

Over the past weeks I’ve worked on a small project to combine the best of the Qt and Python domains. It was time to put both together. I knew about the PyQt- (Riverbank) and PySide- (Qt) bindings for years, but never really dipped my feet into those water. It was time to fix this.

GroundSpace (wordplay) is a small tool to fill your hard-disk (SSD ..) with arbitrary content. To test the speed of writing and to create big chonks of data.

What was learnt?
* creating an ui-file with QtDesigner (jk, I knew this) and how to pre-compile it for PyQt-usage
* loading that uic-file and creating connections
* progress-callback
* how evil the ‘eval()’ function in Python is

Next stop: I want a proper web-scraper in Python.