Curated Resources

Prefer primary documentation and high-signal open-source curricula over random blogs. Links are entry points — always check for updates.

---

Open-source curricula (start here)

Resource	Best for	How to use with this course
DAIR.AI Prompt Engineering Guide · promptingguide.ai	Prompts, CoT, ReAct, agents overview	Modules 01–03, skim before building agents
mlabonne/llm-course	LLM engineer track (RAG, deploy, quant)	Modules 06–07, 09, 13, 17 — take “Engineer” not full “Scientist” unless training is your job
rasbt/LLMs-from-scratch	Attention, GPT training loop	Hybrid track + mental model for 06 / 17
huggingface/agents-course	smolagents, LangGraph, agentic RAG	Modules 11–12, 26, agentic pieces of 09
humanlayer/12-factor-agents	Production agent design principles	Modules 05, 11–13, 20–25, and the orchestration graph
Anthropic / MCP 2026-07-28	Tools, resources, prompts protocol	Module 08 (authoritative, current spec) + host policy in 21

---

Inference performance reading route

Start with Module 28 — Inference serving. Its six lessons include the explanations, worked exercises, and answers. Modules 05, 10, 13, 17, and 24 point at the sections they use. Amit Shekhar's inference article motivated the module; its links are attribution, not required reading.

Application developers should prioritize prefill/decode timing, cache distinctions, and context/concurrency budgets. Self-hosting learners can continue into paged allocation, continuous batching, attention kernels, and speculative decoding. Kernel implementation and GQA uptraining are optional specialist topics.

Models & providers

• OpenAI docs
• Anthropic docs
• Google AI Gemini
• Hugging Face Hub
• Ollama

---

RAG & embeddings

• FAISS
• Chroma
• Qdrant
• Pinecone docs
• Sentence-transformers / provider embedding guides
• LlamaIndex & LangChain retrieval docs (orchestration patterns)

---

Agents & orchestration

• LangGraph
• CrewAI
• AutoGen / AG2
• Provider agent / tool-calling guides (OpenAI, Anthropic)
• Course Module 18 (leaf patterns), 19 (workflow shape), 20–27 (reliability, sandbox, harness, evals, drift, durable graphs, orchestrator comparison) — read 11–12 first
• ReAct (Yao et al., 2022) · ReWOO (Xu et al., 2023) · Self-consistency (Wang et al., 2022)

---

Harness engineering

• Anthropic — Effective harnesses for long-running agents — progress files across context windows
• LangChain — Improving deep agents with harness engineering — same-model / harness-only quality delta
• Course Module 27 — prompt vs context vs harness; src.harness teaching loop

---

MCP

• Model Context Protocol — not a load balancer; tools/resources/prompts over a standard host↔server protocol. Course teaches 2026-07-28 (stateless _meta + server/discover). The 2025 initialize + Mcp-Session-Id model is historical and still deployed.

---

Fine-tuning

• PEFT
• TRL
• QLoRA paper: https://arxiv.org/abs/2305.14314
• Course lab: QLoRA on your own data — adapted from Rahul's small-LLM fine-tuning guide (attribution; the lesson is self-contained)
• vLLM LoRA serving

---

Evals & observability

• promptfoo
• DeepEval
• RAGAS
• Langfuse, Arize Phoenix, OpenTelemetry

---

Security

• OWASP LLM Top 10 (search current version)
• Prompt injection write-ups from major vendors (treat as living docs)
• Microsoft Presidio (PII detection) for production redaction

---

MLOps & serving

• FastAPI
• Docker
• GitHub Actions
• vLLM / Text Generation Inference docs

---

Deep learning (hybrid track)

• PyTorch tutorials
• PyTorch nn.Transformer

---

Classic data / ML

• Pandas 10 minutes
• scikit-learn user guide
• yfinance

---

Course provenance

Pre-restructure originals: archive/source/ in this repository.
Progress (local XP): Progress & gamification.