Chris Mueller

What if software tests could understand themselves? 🤔 AI is already changing how we develop and test software. But that's just the beginning. At this year's Quality Day Berlin, Jonas Menesklou, CEO and co-founder of AskUI, will show how artificial intelligence can be used in quality assurance in practice – and why intelligent test agents are more than just a pipe dream. In his session, Jonas will take participants on a hands-on journey: How can AI-powered agents be developed that automate UI tests, make decisions and adapt test processes independently? Best of all, no prior experience with AI is required – just curiosity and an interest in technology. The presentation is part of a diverse programme on quality intelligence that brings together experts from industry and research. From governance and accessibility to probabilistic models and practical testing strategies, Quality Day promises pure inspiration. 🚀 Be there when quality and AI meet! ➡️ https://lnkd.in/dgkBVXVU organisators: Dr. Erhardt Wunderlich, Sascha Hackel, Heiko Zastrau, Markus Rogawski sponsors: iSQI Group, ITPower Solutions GmbH mediapartner: dpunkt.verlag, GermanTechJobs, SIBB // Digitalverband Berlin-Brandenburg #ASQFEvents #ASQFQualityDayBerlin #DPSQ

7

#FluidXInAction #n8n #Integration #API #Automation #WebRTC #Streaming #LowCode Open. Fast. Integrated. The fluidX THE EYE API documentation has just been updated — and now includes a ready-to-use n8n workflow template for fast integration. With this workflow, you can: • Create a new THE EYE session via the fluidX API • Automatically send an SMS to invite the customer • Automatically send an email to the agent — both connect instantly via browser 🎯 Built as a reference template, this workflow shows how easily THE EYE can be embedded into your own service, support, or customer platforms, portals, or low-code environments. Developers and integrators can adapt and extend it for custom use cases — ticket systems, data pipelines, or automation flows. 👉 Links to the API docs and template are in the first comment. When vision meets code. #FluidXInAction #API #WebRTC #Streaming #Automation #Integration #LowCode #DeveloperTools #fluidX

4

1 Kommentar

How AI anomaly detection protects our Miles & More members!  Our CIO Dominik Sommer and Wibke Laier from Elastic shared fascinating insights at the 2nd Künstliche Intelligenz – Eine F.A.Z. Konferenz about how we leverage the search engine Elasticsearch to detect miles thieves: 🔍 Our AI system analyzes millions of transactions every day   🚨 Suspicious patterns are identified in real-time   ⚡ Fraudulent activities are stopped immediately  🛡️ Our members stay protected What makes us particularly proud?  While many AI projects still end up in the "pilot graveyard," we've already achieved concrete success. Our intelligent anomaly detection isn't just theory – it protects the data  of our 36 million members worldwide every single day. This demonstrates: Successful AI implementation requires clear objectives, clean data, and above all, the courage to move from pilot phase to practice. The future of customer loyalty is intelligent! #MilesAndMore #AI #Innovation #Cybersecurity

138

2 Kommentare

Yesterday, Prior Labs released a new model called Tiny-TSM. It was interesting to see Frank Hutter's team move away from tabular models toward more classical time-series architectures. [Edit: Frank pointed out this was a side project of Felix] 🧐 I’m generally aligned with the idea of small, efficient models, but several methodological choices were surprising, particularly given the claims made. ‼️ Questions for Felix Birkel and Frank Hutter: - As you noted, evaluating on a single dataset and drawing broad conclusions is (at least) problematic. Given that, why did you also drop ~60% of the unique series in Gift-Eval (fin/econ)??? - Excluding them seems hard to justify especially for a model claiming fast, general forecasting performance. 😳 - Gift-Eval uses the geometric mean of MASE as the primary metric. Why did you deviate from this and introduce your own metric? 😬 - Related: you don’t seem to use the geometric averaging required by Gift-Eval when aggregating results. What averaging method did you use, and why not adhere to the benchmark’s methodology? - Why did you omit mentioning the prior work from Vijay Ekambaram from IBM on Tiny Mixers? It is hard to believe you were not aware of that line of research and their great results. - Regarding the drift features: if they effectively triple the input size, wouldn’t a 3× faster convergence simply result from feeding 3× more data per batch rather than architectural advantages? - Finally, should we understand that TabPFN-TS and TempoTabPFN are now outperformed by Tiny-TSM? #HappyForecasting

43

8 Kommentare

🎤 The Human Stack - Vielfalt als Performance-Booster Ever noticed how much we talk about code quality, refactoring, and performance in tech, but not so much about the human stack behind it? In her talk, Franziska Gertz will explore why diversity is the most stable framework we have. Through personal experiences and clear, data-driven insights, she’ll show how real innovation happens when different perspectives, backgrounds, and ways of thinking come together. Thank you so much, Franziska, for bringing this great & relevant topic on our stage 💜 🗣️ Please note: This session will be delivered in German 🇩🇪

10

1 Kommentar

LLMs run everything now... sneaky attacks—like hidden "prompt injections" or "jailbreaks"—can quietly steal data, create harmful software, or break things without anyone noticing. This short guide from Germany's BSI (released November 2025) makes these dangers easy to understand. It uses real-life examples, like planting spyware in an LLM's memory, and shares a step-by-step defense plan. The plan covers training people, checking inputs, limiting what the AI can do, and tweaking the model itself—perfect for developers and security experts. With helpful diagrams, proven design tips, and a simple checklist, it shows how to make your real-world AI setup tougher against threats, without extra hassle. It turns weak spots into strong shields. If you're building or using AI safely, grab this—it's your edge against hackers who trick open models. #datasecutiry #bsi #onlinesecurity #dataprotection #AI_driven_malware

5

Titel: The Great Capitulation: Why Unity’s Lack of Transparency is Failing Developers / Warum Unitys mangelnde Transparenz Entwickler scheitern lässt. I have reached a breaking point with Unity. What we are witnessing is not just technical complexity; it is a total capitulation of the engine’s mission to support creators. The core issues / Die Kernproblematik: 1. The Shader & Pipeline Blackbox: Unity expects us to master complex shader pipelines and 3D texturing for pro-level results, yet provides zero comprehensive documentation on how to actually handle these assets. We are left to guess, while the engine "strips" our work during deployment, leaving us with nothing but a blank skybox on consoles. 2. "Pay-to-Play" Knowledge: It seems as if the industry has collectively decided to gatekeep knowledge. Nobody is willing to show the "how-to" anymore—they only want to collect fees. Tutorials are outdated or intentionally superficial, pushing developers into expensive support channels instead of providing real solutions. 3. Misleading Systematics: Offering console deployment as a "Pro feature" while the technical path exists in the free version is a deceptive practice. It creates an artificial barrier designed to force growth-stage teams into overpriced subscriptions. My position / Mein Standpunkt: This is not "business as usual"—this is a capitulation by engine developers who are no longer willing to educate, only to monetize. It is shameful for the entire gaming industry. If you want us to pay, provide us with the tools and the knowledge to build. Do not hide your lack of optimization behind complex shaders and confusing license structures. Moving forward / Mein nächster Schritt: I am building my own solution: the T7 Game Engine, featuring my Radiance Graphic Engine Pipeline. My goal is true performance and autonomy. I am done with closed pipelines that sabotage the work of indie developers. If Unity’s pricing and transparency do not improve immediately, I will escalate this to consumer protection authorities in both Germany and the EU. This industry thrives on innovation and shared knowledge, not on deceptive marketing and technical sabotage. To those who value transparency and real engineering: Let’s connect. #Unity #GameDev #T7GameEngine #RadiancePipeline #TechIndustry #Transparency #IndieDev #ConsumerRights #GameEngine #ShaderProgramming Unity Xbox Microsoft

1

The german techcrunch is embracing Klaus Programmieren, this is an important step to spread the word about Sovereign AI and fully compliant programming solutions. https://lnkd.in/ey5TCRwA

5

Last week I joined Atlantic and Slush in Berlin to talk about product vision, Industrial AI, and what learnings I made along the journey that has culminated in the founding of Marentis. One lesson has stayed with me: In industrial operations, you always build for two audiences. Executives need a vision strong enough to justify change. Operators need something that works reliably in the real world. You do not unlock one without the other. That belief is now central to what we are building at Marentis with Orhan Can Görür, Dr.-Ing. Can Görür. Marentis is building the operational intelligence layer for industrial machines, turning fragmented machine data into clarity, decisions, and action. Thank you to Atlantic, Bastian Bullmann, Defne Keser, and Slush for the conversation. #IndustrialAI #Marentis

15

My first paper 🎉 Excited to share that Werner Bailer and I will present our work “A Dataset and Metric for Textual Video Content Description” at the ACM Multimedia Conference 2025 in Dublin next week (October 27–31). 🇮🇪 Our work introduces FM-V2T, a new bilingual dataset (English & German) for video-to-text description, along with #LLMFactsF1, an LLM-based metric designed to evaluate the factual accuracy and consistency of generated video captions — automatically. The approach supports applications in media retrieval, multimodal analysis, and natural language processing (NLP). This work was carried out at JOANNEUM RESEARCH in collaboration with the Österreichische Mediathek, within the Austrian #DigitaleTechnologien project #FAIRmedia and the #HorizonEurope project XReco. 📄 Read the paper here: https://lnkd.in/dQthKdxB See you in Dublin! 👋 #ACMMM2025 #AI #LLM #VideoUnderstanding #Research #Multimedia #Dataset #NLP

33

Just spent a day at PyCon Austria 2026 in Eisenstadt — a free, volunteer-run Python conference. Exactly the kind of event that reminds you why the open-source community is worth being part of. The talk that stuck with me the most: "How to Automate Tests for LLMs That Never Answer the Same Way Twice" by Manuel Alejandro Ledezma Falcon. The core problem he described is real and underappreciated: LLMs are non-deterministic by design, so traditional test assertions — built for systems that return the same output every time — simply don't work. His approach using semantic evaluation instead of string matching was practical and immediately applicable. If you're working with LLM-based features in production, this is the kind of thing you need to think about before your test suite becomes a source of noise rather than signal. Coming from a QA and test automation background, this resonated with me. The tooling is catching up, but the mindset shift — from "does it return exactly this?" to "does it mean the same thing?" — is the harder part. #PyCon #PyConAT26 #Python #TestAutomation #LLM #QA #SoftwareTesting

2

Why being a "Team Player" is the Most Valuable Skill in Tech 🤝 Our CEO, Stephanie Levy, is convinced: The most valuable skill our students can learn at 42 Wolfsburg isn't just a programming language—it’s becoming a true team player. In the tech world, there is a simple truth: "You can go faster if you work alone, but you will be stronger if you work together." At 42, this isn’t just a nice-to-have; it’s our core mission. Through our peer-learning methodology, students learn that solving complex challenges requires more than just individual brilliance. It requires: Collective Intelligence: Combining different perspectives to find better solutions. Resilience: Supporting each other when the code gets tough. Growth: Learning that your own success is tied to the success of your peers. We are not just training coders; we are building the collaborative leaders of tomorrow. 🚀 #42Wolfsburg #Leadership #FutureOfWork #Teamwork #PeerLearning #CodingCommunity #TechMission

56

All Code Is "Throw-Away" Code. I was reading the complaint of an engineer recently who was worried that the code she was working on for her project was "Throw-Away" code and wasted effort. It made me reflect on the thousands of lines of code I have written, and that have eventually been deleted or never even used because the project didn't launch. (I do have a personal anecdote of a production bug that lasted 8 years at Amazon below.) I admire the desire to write code that is perfect from the start, but it implies the impossible scenario of knowing everything about the customer, the problem and even the O/S, system and hardware and all data inputs that software will run on in the future. Today, AI tools and ever evolving requirements mean that the most valuable trait is not code perfection, but organizational agility. If you want to get better code: 1. Move Quicker Speed is not about typing faster; it’s about reducing the cost of changing your mind. a) Ship to Learn: The primary purpose of the next feature is to generate a measurable outcome (a key metric, a piece of customer feedback). The moment the feature has taught you what you needed to know, the code has served its purpose. b) Two-Hour Time-Box: If a design decision or refactor takes longer than two hours to implement, you're making the "batch" too big. Break it down. Small changes are easy to revert; large changes slow things down and increase fragility 2. Embrace Ephemeral Code If the code is destined for change surround it with infrastructure to make change easier. a) Test your code: Unit and integration tests are not for the code you write today; they are the scaffolding that allows you to safely rewrite code tomorrow. If you lack tests, your code is functionally permanent because no one can risk changing it. b) Don't Engineer for "Someday": Avoid building abstractions or architectural layers for a requirement you might have in six months. That's speculative engineering. Write the simplest possible code that solves today's problem. If the requirement arrives later, you will have better ability to refactor the simple code into something robust. Anecdote: The Bug That Survived I had a bug that was live from 1998 to 2006. I copy/pasted some code to our Amazon.de search engine and in the process I passed a bad parameter to the bsearch() C-intrinsic, which resulted in a bug that would exhibit differently under different debug flags, optimization flags, OS versions, and system load. The code worked ... mostly .... if the piece of memory allocated happened to be cleared. However, as a bug, it survived longer than most code I have written. (To our German customers: Es tut mir Leid!) If you build code that is so transparent and malleable that if it lasts for years, it’s because it was constantly reviewed and validated, not because it was too scary to touch.

29

5 Kommentare

TYPO3 CMS Migration framework In 2025 I worked on a big migration and upgrade project. It was a website based on TYPO3 CMS, that needed a major technical overhaul. During the upgrade process I migrated over 100 content elements from DCE to Content Blocks and Container. On the TYPO3camp Hamburg 2025 I presented my approach to a handful of people. Some of them approach me afterwards asking about a similar migration for Flux based content elements. Matthias Krappitz put in some effort and found 4 companies including his own to sponsor the development of the Flux to Content Blocks migration. The result is not only a Flux to Content Blocks migration. It is an extensible migration framework for content element migration. The core uses Content Type Providers, that provide normalized configuration and data of the source content elements, Content Type Builders, that build the configuration for the target content type and Record Data Migrators, that take the source data and return the data structure for the new content type. Based on the structure it is possible to migrate from one content element strategy to another. It would even be possible to migrate plugins or core content elements with this, if someone built content type providers for that. Currently there are extension with Content Type Providers for DCE and Flux available, as well as Content Type Builders for Content Blocks and Container: * https://lnkd.in/eg-pirJD: Migrator core * https://lnkd.in/ezGYBXnw: DCE Content Type Provider * https://lnkd.in/eRkUDCPx: Flux Content Type Provider * https://lnkd.in/etgSH_qZ: Content Blocks Content Type Builder * https://lnkd.in/etiDADVs: Container Content Type Builder If you think this could be relevant for you, please start with reading the documentation (README) of the migrator core, as it describes the core concepts and components. If any questions arise please message me here on LinkedIn or send me an email to thorben@webcoast.dk. This the first version, which still needs some testing. So if you're interested, please give it a try an provide feedback. A big thank you goes to aemka, Apart Design Studio, Homepage Helden GmbH and Henrik Ziegenhain for sponsoring the development. Another thank you goes to Nikita Hovratov (https://lnkd.in/efiDk59H) for his work on Content Blocks and b13 GmbH for their container extension. #typo3 #cms #dce #flux #contentblocks #container #sponsor #community #opensource

9

"In other words; the current German EUDI Wallet architecture design appears unable to invoke provision (2) on p. 35 of CIR 2024/2981 (*) avoiding the use of such HSM firmware. This impacts implementation development speed, scalability, cost, and also introduces vendor lock-in."

7

🎵AI roundtable - lively discussion about music and AI together with ChatGPT🤓at the Music School Congress As I have already mentioned, we participated in the music school congress - VdM Congress in Dresden  a few weeks ago and you may have already seen the impressions of the event through my pictures. 🎧Today I would like to share with you a video of the panel discussion, in which I was invited to speak in front of a large audience. 🚀It was a unique experience, with a lively discussion about AI and music. The organiser had the idea of adding ChatGPT 🤖as a discussion partner. 🎵 I was very pleased that so many people were interested in this new topic in the music world. 🤖Also our software, which automatically transcribes music, attracted a lot of interest, particularly with regard to lesson preparation and student motivation. 🎵 Thanks to the organiser and of course to the many listeners in the audience. 🎵A big thank you goes from Klangio to the Verband Deutscher Musikschulen E.V. for the great organization and the varied program. 🚀If you are organising an event that requires expertise in the field of AI and music and you are still looking for an expert / speaker, please contact me and provide some information about your event topics and dates.

9

Last Thursday, I held my first session at the "AI Software Engineering Bootcamp" 🤖 of Digitale Leute School. With AI, the capabilities and responsibilities of engineers are widening. Engineers can - and need to be - closer to the customers and to the product. That is why the first module is about Product Management. 🚀 In my session about "Stakeholder Management and Storytelling", we dove deep into: → Reading the room: How to address stakeholders in different kinds of companies (business-led vs. tech-led vs. product-led) → Prioritizing with spine: Prioritization frameworks that work → Working backwards: Using PR/FAQs to align stakeholders around big bets by turning complex products into simple stories → Saying "no" without burning bridges: How to do your most important job without risking relationships Despite being three hours in the evening, I was very happy about the energy and engagement of the participants. 🔋 My next session is coming up in Feb and will be about "Coding with AI responsibly". Looking forward!

62

8 Kommentare

Wow, yesterday's VDSG meetup surpassed all expectations with around 120 attendees! Here are my key takeaways: Mara Weinblatt Ann-Sophie Brenk Dr. Marian Lux and Markus Stöhr delivered an extensive presentation on AI Factory Austria, the local chapter of the European High Performance Computing Joint Undertaking. Paul Tiwald from Mostly AI showed us how to approach synthetic data generation the right way. I had to cut off Q&A for both talks; otherwise, we'd still be there! :) What I found personally compelling: 🇪🇺 EU AI Sovereignty in action - The European HPC demonstrates Europe's serious commitment to AI independence and infrastructure. I mean actual €€€s 💻 Accessible GPU compute for innovators - Have a compelling use case? Just ask, and AI Factory Austria gives you access to GPU-enabled high-performance computing clusters across the EU. 🏢 Vienna's AI hub is coming - A local Co-Working/Incubator space dedicated to AI companies launches early next year ⚡ Synthetic data made simple – `pip install mostlyai[local]` and you can generate your first synthetic dataset in under 5 minutes. Thanks to everyone who joined us for this Vienna Data Science Group event - hope you had a great time! Special thanks go to Florian Thürk for putting this event together, Wolfgang Weidinger and Georg Heiler for all the support, Andreas Nemeth and Cloudera for sponsoring and to René-Marcel Wiese and Claudia Krügel for one of the most professional ops we've ever had. We are hoping to see you at our next one! #AI #ArtificialIntelligence #MachineLearning #Vienna #VDSG #AIFactoryAustria #SyntheticData

123

4 Kommentare

🎙️ Podcast weekend edition: Moving from "AI as a helper" to "AI as a doer" – Exploring Agentic Coding with Straion 💻🚀 Is the way we build software being rewritten in real-time? In the latest episode of AI Talk, Jakob and I are thrilled to welcome Katrin Freihofner, founder of the Linz-based startup Straion. Having recently closed a funding round, Katrin and her team are diving deep into one of the most dynamic fields in tech: Agentic Coding. In this episode, we discussed: 👩‍💻 Beyond Copilot: Why Straion focuses on the "agentic" part of development and how it differs from the tools we’ve seen so far. ⛰ The DACH Startup Edge: How a lean, focused team from Austria can compete in a global market dominated by Silicon Valley giants. 🔦 The "Flashlight App" Problem: Will specialized software tools survive, or will general-purpose agents eventually absorb them all? 🤝 Future-Proofing Talent: Why human empathy, strategic thinking, and "common ground" remain the ultimate primitives of success, even as the code writes itself. I’m already looking forward to our follow-up talk for their Series B next year. /jk Their speed of execution in this space is breathtaking. Watch the full interview and read our deep-dive article here: 📺 https://lnkd.in/d3vs2_J4 Lukas Holzer Matthias Seiderer Andreas Greilhuber Fabian Friedl Trending Topics by newsrooms newsrooms Adam Ehrlich Matthias

43

1 Kommentar