Book notes · Chip Huyen

AI Engineering

A decision framework rather than a tutorial: how to pick models, techniques, data strategies and deployment patterns, with the book's figures, architecture diagrams, key takeaways and a practitioner checklist per chapter.

10 chapters about 6.5 hours end to end 152 figures

Start reading

Chapter 1 29 min 17Introduction to Building AI Applications with Foundation ModelsHow language models became foundation models, the eight use-case families, planning an AI product, the three-layer AI stack and AI vs ML engineeringRead
Chapter 2 39 min 24Understanding Foundation ModelsTraining data, transformer attention and the KV cache, scaling laws, post-training with SFT, RLHF and DPO, sampling settings, and why models are inconsistent and hallucinateRead
Chapter 3 35 min 9Evaluation MethodologyWhy evaluating foundation models is hard, perplexity and its relatives, exact and similarity scoring, embeddings, AI judges and their biases, and Elo-style model rankingRead
Chapter 4 38 min 9Evaluate AI SystemsEvaluation criteria, the model selection workflow, open versus proprietary models, public benchmarks and contamination, and how to design an evaluation pipelineRead
Chapter 5 38 min 16Prompt EngineeringIn-context learning, prompt anatomy, system prompts and context length, prompt best practices and versioning, and defensive prompting against extraction, jailbreaks and injectionRead
Chapter 6 44 min 16RAG and AgentsRAG architecture and retrieval algorithms, vector search and retrieval optimization, agents with tools and planning, agent failure modes, and memoryRead
Chapter 7 49 min 18FinetuningWhen finetuning beats prompting and RAG, the memory math that makes it expensive, PEFT and LoRA, quantization, model merging, and the tactics that make a run succeedRead
Chapter 8 36 min 8Dataset EngineeringCurating post-training data for quality, coverage and quantity, synthesizing and verifying data with AI, distillation, and the inspect-dedupe-clean-format pipelineRead
Chapter 9 49 min 17Inference OptimizationCompute versus memory bottlenecks, latency and throughput metrics, accelerators, and the model-level and service-level techniques that make LLM serving fast and cheapRead
Chapter 10 44 min 18AI Engineering Architecture and User FeedbackGrowing an AI application from a bare model call to a guarded, routed, cached and observable pipeline, and turning user behaviour into feedback that improves itRead

AI Engineering

Search

AI Engineering