Chip Huyen

Chip Huyen is co-founder of Claypot AI (real-time ML); former NVIDIA (NeMo platform); former Netflix AI researcher; Stanford machine learning lecturer; author of AI Engineering (most-read O’Reilly book since launch) and Designing Machine Learning Systems. Works with enterprises on AI strategy and product development.

Key ideas

• Talk to users, not AI news. The most common mistake in AI product development is investing in infrastructure and model selection when the high-leverage work is user research, data preparation, and prompt engineering.
• Data preparation is the real RAG work. Chunking strategy, contextual metadata, and hypothetical question generation drive RAG quality more than infrastructure choices.
• Evals are an ROI question. Always valuable in principle; the real decision is opportunity cost and expected gain relative to alternatives.
• AI Engineering as a profession. Using pre-trained models to build applications is a distinct discipline from ML Engineering (training models). Lower entry barrier, high ceiling.
• Test-time compute. Allocating more inference compute (multiple samples, extended reasoning) substantially improves perceived model quality without changing the base model.

Appearances

See also

• AI Engineering
• Evals
• Reinforcement Learning from Human Feedback
• Scaling Laws