九章完整路径 — 每章配套 16:9 幻灯片 + 竖版 scroll 文档，内容一致。
Transformer, inference, sampling, Pretrain / SFT / RLHF, model selection.
Problem-first: hallucination, stale data, private docs. Look up first, then think.
Why LLM ≠ Agent. ReAct loop, tool calling, intent routing, multi-turn context.
BM25, dense, hybrid, rerank, model selection, RAG eval metrics.
Framework comparison first, then LangChain RAG chain, LangGraph ReAct, SqliteSaver + Store.
Model Context Protocol (tools), build MCP server, Agent-to-Agent handoffs.
Nous Hermes tool-calling models, OpenClaw agent OS, Skills, ClawHub, self-host lab.
Supervisor, handoff, critic patterns. GraphRAG vs vector RAG.
ViT, SAM, Stable Diffusion. Whisper ASR. Neural TTS.
Cross-chapter metrics, golden-set workflow, per-chapter eval sections, debug decision tree.
serve-present.bat on PC, open http://PC_IP:8080/present-hub.html in Safari.