Self-study notes from Chip Huyen's AI Engineering (O'Reilly, 2025), chapters 1–10.
I read this to sharpen the fundamentals behind production LLM / agentic systems — evaluation, RAG & retrieval, prompt engineering, dataset engineering, inference optimization, and end-to-end AI architecture. For each session I read a couple of chapters, wrote up the core concepts in my own words, dug into topics the book only touched on, and logged the questions I still had.
📝 The detailed chapter notes are written in Korean (my study language). This README summarizes them in English.
I distilled this study into an internal company talk: "Dataset Engineering for LLM: 학습 단계별 데이터 설계" (Dataset Engineering for LLM — designing data across training stages).
| Chapters | Topics | Notes |
|---|---|---|
| 1–2 | Foundation models, prompting (CoT, n-shot, structured output), decoding (beam search), preference fine-tuning (DPO vs RLHF, RLAIF), multi-head attention, non-transformer architectures (RWKV, SSM) | notes/chapter-1-2.md |
| 3–4 | Evaluation: RAG evaluation with RAGAS (faithfulness / answer relevancy / context recall & precision), LangSmith evaluation workflow, G-Eval & LLM-as-judge | notes/chapter-3-4.md |
| 5–6 | Prompt engineering (in-context learning as Bayesian inference, persona trade-offs, prompt versioning, DSPy) + RAG retrieval (vector indexing — Flat / IVF / HNSW, FAISS, ANN), intent routing | notes/chapter-5-6.md |
| 7–8 | Dataset engineering & cost: prompt caching, quantization (1.58-bit), data-centric AI pipeline, data flywheel | notes/chapter-7-8.md |
| 9–10 | Inference optimization (TTFT/TPOT, MFU/MBU, speculative & parallel decoding, KV cache, batching, prefill/decode disaggregation) + AI engineering architecture (context, guardrails, router/gateway, caching, agent patterns, observability, orchestration) + user-feedback systems | notes/chapter-9-10.md |
.
├── README.md # this overview
├── notes/ # chapter notes (Markdown, Korean) — concepts · reflections · open questions
│ ├── chapter-1-2.md
│ ├── chapter-3-4.md
│ ├── chapter-5-6.md
│ ├── chapter-7-8.md
│ ├── chapter-9-10.md
│ └── images/
└── handwritten/ # scanned handwritten notes (PDF), ch.1–8