#59 Carsten Kraus - Industrial AI, keep the data in-house
Carsten Kraus's pitch for the German Mittelstand isn't to build foundation models. It's to keep the data in-house and let other people's models run on it. In this episode he lays out why that matters more than chasing the next LLM, and what European companies should actually do about AI right now. Kraus has founded 12 companies and is opening the 13th. He sold a programming language called Omikron BASIC to Atari at 17, built FactFinder (similarity search two years before Google's "Did you mean", market leader for a decade), and Casablanca.ai, a face foundation model that hit 88% AUC on stroke detection in five days with only 150 training images. Germany's Business Angel of the Year 2024, Forbes Technology Council, 25 patents. Lives and works in Pforzheim, not Munich or Berlin, and explains why that matters. We cover why Aleph Alpha lost on capital and compute rather than technology, how DeepSeek and Kimi cut training cost, what Qwen 3 and a single 4090 can actually do for a mid-sized company running local AI, and how worldview gets baked into training data, including the conspiracy-theory fine-tune Carsten's team ran but never published. Then: where the European Mittelstand has a real edge (industrial AI on top of 1,800 hidden champions), the hyperscaler problem (the US-East-1 outage, the UK tax software that ran in the wrong data center, CLOUD Act exposure), and where to start if you're a CEO or CTO in a German mid-sized company that hasn't moved yet. Chapters: 0:00 Why this conversation: the state of AI in the German Mittelstand 1:36 Atari at 17, then FactFinder: Carsten Kraus's invention track record 11:02 Casablanca AI: a face foundation model that beat universities at stroke detection 14:17 Why he didn't leave for Silicon Valley 18:16 Talent flight: Black Forest Labs and five friends who left Germany 20:40 The actual European lever in AI (it's not foundation models) 24:23 Aleph Alpha post-mortem: compute, capital, and what Europe missed 26:45 How DeepSeek made training cheap, and what Qwen 3 27B runs on today 30:00 RAG on a 4090: smaller models with bigger context windows win 33:10 Worldview in training data: death penalty, contracts, and the m-dash tell 40:46 Industrial AI, the 1,800 hidden champions, and keeping data in-house 43:25 Where the Mittelstand should start, and the hyperscaler trap 46:48 Kraus's next project: capturing departing employees' knowledge with AI About the guest: Carsten Kraus is a serial founder, AI researcher, and Forbes Technology Council member based in Pforzheim. He holds 25 patents under his own name and was named Germany's Business Angel of the Year 2024 by BAND. His companies include FactFinder (e-commerce product search, ~1,800 customers including Lidl, OBI, Stihl, Mytheresa), Casablanca.ai (eye-contact correction for video calls), and Brainbox, the AI incubator inside his CK Holding group. Links & resources: Carsten Kraus on LinkedIn: / carstenkraus Casablanca.ai: https://www.casablanca.ai CK Holding (parent of Brainbox, Omikron, Casablanca): https://www.ck-holding.de FactFinder: https://www.fact-finder.com Omikron BASIC on Wikipedia: https://de.wikipedia.org/wiki/Omikron... Business Angel of the Year 2024 (BAND): https://www.business-angels.de/en/hea... Black Forest Labs (Freiburg, now also San Francisco): https://bfl.ai Aleph Alpha: https://aleph-alpha.com DeepSeek: https://www.deepseek.com Kimi (Moonshot AI): https://www.moonshot.ai Qwen (Alibaba open models): https://qwenlm.github.io Common Crawl: https://commoncrawl.org StackIT (Schwarz Digits cloud, Lübbenau data center): https://www.stackit.de ━━━━━━━━━━━━━━━━━━━━ The Stellar Work Podcast Conversations on transformation, flow, and the future of work. 🎧 Listen on all platforms: https://stellarwork.start.page 📬 Newsletter: https://substack.com/@stellarwork 💼 Host: Ben, founder of Stellar Work ━━━━━━━━━━━━━━━━━━━━

The most rational take on AI you’ll hear this year

Yann LeCun Says LLMs Have 2 Years Left…

Engineering Management In The Age Of AI

Something is jamming GPS over Europe. Here's what we found

#55 Allen Holub About The Evolution Of Agility And Its AI Future

Zig 2026: No-AI Policy, $670K Foundation, Left GitHub & Why Zig Isn’t 1.0 - Andrew Kelley Explains

Trump Preps for 80th Birthday, Threatens to Hit Iran, Knicks Historic Win & Elon Musk Trillionaire!?

Turing Award Winner: Disagreeing with Google, Postgres, Future Problems | Mike Stonebraker

Demis Hassabis: Why AGI is Bigger than the Industrial Revolution & Where Are The Bottlenecks in AI

From Idea to $650M Exit: Lessons in Building AI Startups

Anthopic, OpenAI Should Not Be Allowed to IPO, Says Ed Zitron
![Yann LeCun's $1B Bet Against LLMs [Part 1]](https://i.ytimg.com/vi/kYkIdXwW2AE/hqdefault.jpg?sqp=-oaymwEjCNACELwBSFryq4qpAxUIARUAAAAAGAElAADIQj0AgKJDeAE=&rs=AOn4CLDbV4izF3i-wxevCVIn7FJjoy1vlA)
Yann LeCun's $1B Bet Against LLMs [Part 1]

Andrej Karpathy: From Vibe Coding to Agentic Engineering w/ Stephanie Zhan

Exclusive Interview With Nvidia CEO Jensen Huang (Full Special)

The future of intelligence | Demis Hassabis (Co-founder and CEO of DeepMind)

Anthropic's Boris Cherny: Why Coding Is Solved, and What Comes Next

FULL DISCUSSION: Google's Demis Hassabis, Anthropic's Dario Amodei Debate the World After AGI | AI1G

This is not the AI we were promised | The Royal Society

Inside YC's AI Playbook

