Recently we have ๐๐ฎ๐ญ๐จ๐ฆ๐๐ญ๐๐ the ๐ซ๐๐ฏ๐ข๐๐ฐ to save everyone’s time.
The agents now review each otherโs fixes and suggestions, the CI queue has its own postcode, and the cloud bill is the only thing scaling faster than the code.
The good news is that the tests still take a few minutes.
Once you get a runner. And of course, this is all fictional. ๐
video:
#
Usually I receive a lot of messages on LinkedIn, which are connect requests, messages from recruiters, and of course the “quick call about your lead generation?” offers. Answering all of them in a friendly and individual way takes time. So I thought about ๐๐ฎ๐ญ๐จ๐ฆ๐๐ญ๐ข๐ง๐ it .. at least partly.
But before writing code: is this even allowed and legal? As techie I’m always wondering how influencers and their teams handle this. I highly doubt they type every comment and react to every message manually. So I gathered information about the terms of use, APIs, lawsuits and user reports. Quite a rabbit hole. Result: LinkedIn is ๐ฉ๐ข๐๐ค๐ฒ about it.
The short summary in HTTP status codes:
* ๐๐๐: means posting via the API works
* ๐๐๐: DMs are only for approved partners
* ๐๐๐: LinkedIn’s own status code for bots
* ๐๐๐ and ๐๐๐: stand for lawsuits against scrapers and removed company pages of tool vendors (2025 & 2026)
So if I were trying to build something like this, then like: a local tool reads via screenscraping and prepares the drafts and sending stays my job.
Of course, this is no legal advice ๐ And if you know more, tell me.
๐ซ ๐๐ก๐จ๐ซ๐๐ฑ๐๐๐ ๐๐ซ๐จ๐ฆ ๐๐ข๐ฌ๐ก
A very good friend of mine had a CT of his chest taken last week. And then he had his images. And no medical evaluation.
So I said: Come, let’s get this done at home and prepare an app where ๐ฐ๐ ๐๐๐ง ๐๐ฏ๐๐ฅ๐ฎ๐๐ญ๐ ๐ญ๐ก๐ ๐ฌ๐๐๐ง๐ฌ ๐จ๐ฎ๐ซ๐ฌ๐๐ฅ๐ฏ๐๐ฌ. The idea was simple: let’s figure out what existing models and tools from the public domain can be used to evaluate those images and detect issues. As well as: how to run them and what quirks are known. After realizing that several of them are not just ML models, but LLMs, the path with local hosting via #Ollama was clear: who would like to see their most intimate data in the cloud?
A bit of #Python for the backend, a bit of HTML and CSS for the front, tons of Ollama models and several iterations later to understand what the models actually return, jobs for queuing work, polishing, done.
๐ I have to admit, if it was this easy, I would have finished this “oh, this should not take long” project way earlier. Several assumptions were wrong, outcomes from random sampling of the CT scans were not good, model quality was … let’s say most are overconfident when praising what they can do – so I also added a pre-rating for each of them. Since I am no doctor – the user is still free to override this. My recommendation is “๐ฆ๐จ๐ง๐๐ข_๐ฅ๐ฎ๐ง๐ _๐ง๐จ๐๐ฎ๐ฅ๐” as a model. Another obstacle was finding good medical data, since I wanted to keep the data to analyse separate from the “development data”. Then: does it make sense to prepare synthetic data (…not) and why is #DICOM a standard when there are some deviations in the medical data?
Nevertheless: all is solved now. The app works and gives human-readable results.
There is still no evaluation from an expert, but based on the app I prepared, I would say: ๐ก๐ข๐ฌ ๐๐๐ญ๐ ๐ฅ๐จ๐จ๐ค๐ฌ ๐ ๐จ๐จ๐, ๐ง๐จ๐ญ๐ก๐ข๐ง๐ ๐ซ๐๐๐ฅ๐ฅ๐ฒ ๐ญ๐จ ๐ฐ๐จ๐ซ๐ซ๐ฒ ๐๐๐จ๐ฎ๐ญ. This is, of course, not a medical diagnosis.
So, mission done.
๐๐ข๐ญ๐๐ฎ๐ -> ๐ฆ๐๐ซ๐๐๐ฅ๐ฉ๐๐ญ๐ซ๐ข๐๐ค -> ๐๐ก๐จ๐ซ๐๐ฑ๐๐๐_๐๐ซ๐จ๐ฆ_๐๐ข๐ฌ๐ก
https://github.com/marcelpetrick/ThoraxGPT_from_Wish
Hacktoberfest 2022 โ 2026: ๐ญ๐ก๐ ๐ฐ๐จ๐ซ๐ฅ๐’๐ฌ ๐ฌ๐ฅ๐จ๐ฐ๐๐ฌ๐ญ ๐ฉ๐ซ๐จ๐ญ๐จ๐ญ๐ฒ๐ฉ๐ now works
October was always some kind of #Hacktoberfest. Four years ago I joined a hackathon with Shuhong and Tom, where we tried to get any of the provided brain-computer interfaces to work. It was nice that such an event was organized by the @TUM. I guess there is also a really old post about this here on LinkedIn.
Anyway: at that time we spent most of the time configuring and getting familiar with the hardware and the whole workflow, instead of building a final prototype. So around noon on Sunday we decided to capture some “๐๐จ๐ง๐ญ๐ซ๐จ๐ฅ ๐๐๐”. I never got around to processing them and testing them against the idea of controlling ๐๐๐๐ฆ๐๐ง with them. But since I came across this GitHub repo while cleaning up, I thought: ๐ฐ๐ก๐ฒ ๐ง๐จ๐ญ ๐ง๐จ๐ฐ?
So, here it is. One stream contained ten blink signals (with the eyes), which were recorded by the #OpenBCI Ganglion, the BCI (brain computer interface) board we used at that time.
They are recognizable by a filter: a 0.5โ10 Hz band-pass, averaged over the four channels, then a robust z-score threshold (median/MAD) that picks the big peaks. Then those impulses are used to control Pacman. As you can see, the result is not optimal and not suited to really “play” that one-dimensional game. But it shows the whole end-to-end workflow.
Now updated on GitHub as a repo, with the tooling, the recorded test data and also the end-to-end connection: GitHub -> marcelpetrick -> NTXHackathon22
https://github.com/marcelpetrick/NTXHackathon22
If you have question, ask. Of course, this is not Neuralink ๐
Some fun for the evening: ๐๐๐ฏ๐๐ซ๐ข๐๐ง ๐๐๐ฆ๐ฆ๐ข๐ง๐ ๐ฌ ๐๐ค๐
Yesterday I had some tokens left (#tokenAnxiety) in some subscriptions, so why not revive a classic from the 90s: Lemmings?
Of course, with ๐๐๐๐๐ซ๐ก๐จ๐ฌ๐ง and ๐๐ข๐ซ๐ง๐๐ฅ, ๐๐จ๐๐ซ๐ข๐ฌ๐๐ก for the menu and fitting folkloristic tunes. The remembered game mechanics are the same.
Give it a try (zero cost, no installation, all in the browser): ๐๐ช๐ต๐๐ถ๐ฃ -> ๐ฎ๐ข๐ณ๐ค๐ฆ๐ญ๐ฑ๐ฆ๐ต๐ณ๐ช๐ค๐ฌ -> ๐๐ข๐ท๐ข๐ณ๐ช๐ข๐ฏ๐๐ฆ๐ฎ๐ฎ๐ช๐ฏ๐จ๐ด
https://marcelpetrick.github.io/BavarianLemmings/
Also has a touchscreen mode! ๐๐ค๐
Video; (open as post)
๐ Individualsoftware fรผr alle ๐ฅณ
Am Wochenende berichtete meine liebe Frau, dass einige Klassen und Tagesheimgruppen ๐รค๐ซ๐ฆ๐๐ฆ๐ฉ๐๐ฅ๐ง einsetzen. Klingt gut, dachte ich. Dieses technische Hilfsmittel kann helfen, alle zu ermutigen, zur Hausaufgabenzeit ruhiger zu sein.
Dann aber: Das ist nur eine manuell zu pflegende Darstellung einer Ampel. Keinerlei Sensorik, etc. ๐
Also schnell Ideen fรผr ๐๐ ๐ฎ๐ง๐ ๐๐ festgelegt, Niedlichkeitsfaktor (wir brauchen ๐๐๐ญ๐ณ๐๐ง) inklusive, etwas ๐๐๐ฆ๐ข๐๐ข๐๐๐ญ๐ข๐จ๐ง und ab ging es.
Was haben wir heute mit ๐๐๐ฎ๐ฆ๐๐ซ๐๐ฎ๐ง๐ ๐๐๐ซ๐ฌ๐ข๐จ๐ง ๐.๐.๐?
Eine vollkommen lokal nutzbare, kostenlose Android-App auf Flutter-Basis. Eine Katze namens Mia, welche bei zu viel Lรคrm temporรคr verschwindet โ kam sehr gut bei den Kindern an. Einstellbare Gerรคuschpegel, Kalibrierungsmรถglichkeit, die Mรถglichkeit, Sterne zu sammeln (Gamification) und damit dann Accessoires fรผr die Katze zu kaufen. Bunte Farben ๐ฆ
Ah, ach so: Es ist leicht, die Katze zu vergraulen, aber schwer, wieder ihr Vertrauen zu gewinnen (Hysterese).
Holt euch ein Release-APK auf ๐๐ข๐ญ๐๐ฎ๐: ๐ฆ๐๐ซ๐๐๐ฅ๐๐ฉ๐ญ๐ซ๐ข๐๐ค -> ๐๐๐ฎ๐ฆ๐๐ซ๐๐ฎ๐ง๐
https://github.com/marcelpetrick/Raumfreund/releases/
#agenticSoftwareEngineering
๐๐จ ๐ฒ๐จ๐ฎ ๐ฐ๐๐ง๐ญ ๐ญ๐จ ๐ค๐ง๐จ๐ฐ ๐ฐ๐ก๐๐ซ๐ ๐ ๐ฐ๐๐ฅ๐ค ๐๐ฎ๐ซ๐ข๐ง๐ ๐ฅ๐ฎ๐ง๐๐ก?
I’ve always wanted to have a tool to display all my activity tracks from ๐๐๐ซ๐ฆ๐ข๐ง on a global map. Which allows zooming and panning, selecting different time spans, etc. So, I created it in June. A minimal viable prototype was quickly done. Issues with broken tracks (GPX and the jitters ..) were fixed, and some optimizations to the loading and caching of the ๐๐ฉ๐๐ง๐๐ญ๐ซ๐๐๐ญ๐๐๐ฉ tiles were also done.
But then the ๐ฌ๐ก๐จ๐ฐ๐ฌ๐ญ๐จ๐ฉ๐ฉ๐๐ซ: it does not scale! I had used some hundred exported tracks initially: loads quite fast and also the mapping to the global coordinates was done within milliseconds. But when I tried to load all of the almost 10k tracks (I have been doing this for almost a decade ..), the app froze!
A classic in software development: it does not scale. What now?
Quite simple: profile a run, determine the bottlenecks and optimize. Of course, part of that can be done in an agentic way. Urging the agents to pre-process all tracks and store them in a more suitable format (QPainterPath) for quick lookup got us halfway there.
So here is version 0.0110 of ๐๐๐ซ๐ฆ๐ข๐ง๐๐๐ญ๐ข๐ฏ๐ข๐ญ๐ฒ๐๐๐ฉ (quite an uncreative name ..) – a ๐๐ฒ๐๐ญ ๐.๐๐ app, with Python 3.14, 11k LoC (4k for the functionality and the rest for tooling & tests), ๐๐% ๐๐จ๐ฏ๐๐ซ๐๐ ๐, local and GitHub Actions pipelines, .. you know the drill. You also receive a nice shell script to export your data from Garmin (if you don’t want to request a GDPR export and wait some days).
๐ ๐ฎ๐ฅ๐ฅ๐ฒ ๐ฅ๐จ๐๐๐ฅ, nothing is sent to the cloud or leaves your computer.
It really stands on the shoulders of giants: big thanks to the people behind OpenStreetMap and all the frameworks used. It would not be possible to create such a custom project without their work.
PS. to resolve the headline: those are the streets adjacent to the DATA MODUL AG headquarters in Munich/Laim.
๐ข Die รra der ๐๐ง๐๐ข๐ฏ๐ข๐๐ฎ๐๐ฅ๐ฌ๐จ๐๐ญ๐ฐ๐๐ซ๐ beginnt am Kรผchentisch
Tobias Boehm hat vor vier Tagen eine klasse ๐๐๐ซ๐ง๐๐ฉ๐ฉ fรผr seine Tochter verรถffentlicht. Bunt, ansprechend gestylt, direkt zugeschnitten fรผr die Zielgruppe: eine Person. Fรผr ein einzelnes Ziel: Sicherstellung guter Noten. (Was ist eigentlich rausgekommen?)
Der Post sprach mich direkt an, da ich in der Vergangenheit ebenfalls versuchte, abseits bestehender Lernsoftware etwas fรผr die Kinder zu erstellen: manchmal mit mehr Publikumserfolg, meist mit weniger. ๐๐๐ฆ๐ข๐๐ข๐๐๐ญ๐ข๐จ๐ง ist definitiv der richtige Ansatz, um die jรผngere Zielgruppe zu interessieren. Bei mir schwang auch der Aspekt mit: Zeig den Kindern einmal direkt, was mit etwas Einsatz, Kreativitรคt und Programmierkenntnissen mรถglich ist. Vielleicht zieht es sie ebenfalls in die richtige Richtung. Vielleicht ist man auch manchmal รผberinvestiert in den Erfolg der Kinder – das mag sein. Aber warum auch nicht?
Aber zurรผck zum Thema: Ich finde das Projekt richtig gut und es zeigt, dass man mit agentischer Unterstรผtzung und dem passenden Software-Know-how individuell zugeschnittene Software-Produkte erstellen kann. Hochqualitativ (siehe Tests und Pipelines) und in kurzer Zeit. Dass es als ๐๐ฉ๐๐ง ๐๐จ๐ฎ๐ซ๐๐ mit passender Lizenz verรถffentlicht wurde, erlaubte mir ebenfalls, einen winzigen Teil beizutragen – direkt nach Review durch Agenten gemergt. Super.
๐ฆ ๐๐๐ญ๐ก๐๐ฅ๐๐ซ๐ง๐๐ง ๐๐๐ซ๐ ๐๐ฉ๐ร ๐ฆ๐๐๐ก๐๐ง! ๐
Link zum Ursprung: https://lnkd.in/p/eRMQqGp9
๐ฅ๐ฅ๐ฅ Hiermit ergรคnze ich meine Qualifikationen um die erneuerte Zertifizierung als Brandschutzhelfer. ๐๐ข๐๐ก๐ญ ๐๐ฅ๐ฅ๐๐ฌ ๐ข๐ฌ๐ญ ๐๐จ๐๐ญ๐ฐ๐๐ซ๐ ;)
Ein wichtiges Thema, auch wenn wir uns natรผrlich glรผcklich schรคtzen kรถnnen, dass die heutigen Sicherheitsstandards das Brandrisiko deutlich verringern.
Vielen Dank an DMO fรผr die Organisation und an Veso Ilic fรผr die Schulung!
Porting to ๐๐ญ ๐.๐๐ – obey the ๐๐๐
At the end of September, Qt 6.12 was released into the wild. And since this is a much-anticipated LTS version (the last one was 6.8), I started to port my FOSS projects.
๐๐ฎ๐ฅ๐ฅ๐๐ง๐๐ฎ๐ฅ๐, since it is a tiny project but quite old (~8 years), went through a lot of changes. It is a helper to cull unwanted photos from a weekend session (aka: just filter the best ones out). Boring old-fashioned widgets with some C++ backend code: read the directory, load image files, move them to different directories, some UI themes and ๐ข๐๐๐ง, nothing fancy.
But that makes it an ideal candidate to try out new Qt releases.
The work was done in less than 30 minutes, thanks to guardrails like 98% test coverage and CI pipelines. This was supported by the use of modern dependencies like ๐๐๐๐ค๐ ๐.๐.๐, ๐++๐๐, etc., which prevented me from having to modernize not just Qt, but other things as well.
So what do we get now? An LTS release, which promises ๐ ๐ฒ๐๐๐ซ๐ฌ ๐จ๐ ๐ฌ๐ฎ๐ฉ๐ฉ๐จ๐ซ๐ญ, which means we should be well defended against cybersecurity vulnerabilities and software flaws (CVEs). Which is part of the CRA. To resolve the pending question what does CRA even mean: this is the ๐๐ฒ๐๐๐ซ ๐๐๐ฌ๐ข๐ฅ๐ข๐๐ง๐๐ ๐๐๐ญ, where the first major enforcement milestone took effect on 11th September 2026. Vendors of software-products are responsible to deliver maintenance
Again, two releases for Windows (๐๐ฑ๐๐๐ฎ๐ญ๐๐๐ฅ๐) and Linux (๐๐ฉ๐ฉ๐๐ฆ๐๐ ๐).
And a resulting ๐ฆ๐ข๐ ๐ซ๐๐ญ๐ข๐จ๐ง ๐ ๐ฎ๐ข๐๐, which is actually the more informative part for me – because I like to know what we have to change to ๐ฉ๐จ๐ซ๐ญ ๐๐ฑ๐ข๐ฌ๐ญ๐ข๐ง๐ ๐ฌ๐จ๐๐ญ๐ฐ๐๐ซ๐.
Fork the 2.6k LoC and 2.7k lines of tests at GitHub: https://github.com/marcelpetrick/Cullendula/
