Industrial AI Podcast Cracking the Code: Scaling AI at Audi
10 snips
Feb 25, 2026 André Sagodi, a manufacturing and innovation lead at Audi with experience scaling AI-based computer vision in production. He discusses scaling crack-detection from pilot to global rollout. Topics include system architecture for one-model approaches, data and labeling strategies, edge vs cloud deployment, and organizational moves needed to turn pilots into production-ready solutions.
AI Snips
Chapters
Transcript
Episode notes
Scaling Is Organizational Capability
- Scaling AI is an organizational capability, not just a technical task.
- André Sagodi contrasts innovation (experimenting and learning) with scaling (replicating proven solutions across sites).
Crack Detection Became A Viral Use Case
- Audi tackled crack detection in deep-drawn sheet metal parts as a high-impact, high-volume use case.
- Press shops produce millions of parts; even ~1/1000 defect rate becomes substantial and drove adoption of AI vision.
One Model To Avoid Model Zoo
- Audi built one universal AI model instead of a model per site to avoid a fragmented model zoo.
- They centralized labeling with a shared annotation guide, consensus labeling and a defect reference catalog to ensure consistency.