Principles for Designing Generative AI Powered Products
summary
Zalando builds a GPT-powered product called Zalando assistant that integrates large language models into search to let customers describe items in full sentences, serving 50 million customers across 25 markets. Teams validate prototypes with in-depth interviews, expose the assistant as a beta, and log failures — hallucinations, context loss and keyword confusion — into an internal learning platform. They ship an intent classifier to block harmful queries and only surface the assistant when it’s meaningful, iterating fast on real-world signals.
AI-written summary · about
chapters (9)
- 0:00Introduction and personal journey
- 1:10Evolution of human-computer interaction
- 2:34Generative AI in e-commerce context
- 5:14Introducing Zalando Assistant project
- 6:00Principle 1: Define unique challenges
- 7:45Principle 2: Guide users through paradigm shift
- 10:58Principle 3: Manage expectations and limits
- 12:24Principle 4: Learn to fail gracefully
- 16:02Summary and closing remarks
related talks
27mBuilding AI applications: Successes and Lessons
29mTransformation in a New Era of AI
14mGenerative AI models: Practical Business Applications
33mAI Design for Good: Leading with Vision & Humanity in the Age of Intelligent Machines
29mImpact of #GenerativeAI in Development and #SoftwareTesting
49mGenerative AI: Concepts, Tools, and Applications for Data Professionals
26mAI-first frontend architecture
30mBuilding generative AI-powered products