AI Engineering: Building Applications with Foundation Models
I keep this at #9 because a technically serious operating manual for designing, evaluating, and deploying foundation-model products. I still need its pressure.
The long version
Chip Huyen maps the engineering discipline that emerges when foundation models become components inside real products. She covers use-case framing, evaluation, prompting, retrieval, agents, fine-tuning, latency, cost, feedback, observability, and the difficult gap between a convincing prototype and a reliable system used by actual people. It reminds me that the way intelligence is becoming a tool, a product, and a decision I still have to own does not happen only to other people.
Why it is here
This is the shelf’s most complete bridge from model capability to production responsibility. It gives you a vocabulary for making architecture and product decisions without confusing demo quality, benchmark scores, and user value. I do not want the lesson to become another sentence I agree with and leave behind.
How to read it
Study it beside a small product you can actually build. For every major chapter, add one decision to an engineering memo: desired behavior, evaluation set, failure budget, latency target, cost ceiling, and escalation path. I will stop long enough to write what the chapter changes, if anything.
What it taught me
- 01
Evaluation must combine model, system, and product-level signals.
- 02
RAG, fine-tuning, and prompt design solve different failure modes and carry different costs.
- 03
Production AI requires feedback loops, guardrails, observability, and graceful failure.
Before and after
What makes it easier
Hands-On Large Language Models
Jay Alammar; Maarten Grootendorst
Designing Machine Learning Systems
Chip Huyen
More about the author
Chip Huyen is a computer scientist, author, and builder of machine-learning infrastructure. Her work focuses on the design and operation of real-world ML and foundation-model systems, with particular attention to evaluation and production trade-offs.
More by the same hand
Designing Machine Learning Systems
Chip Huyen
Machine Learning Interviews
Chip Huyen