Stack

Understand

How the models actually work, from the tokenizer up.

• Tokenization, Inside and Out · 14 lessonsBPE, WordPiece, Unigram, SentencePiece, tiktoken — every algorithm, every tradeoff, every production implication.
• Attention Mechanisms, from First Principles · 16 lessonsSelf-attention, multi-head, KV cache, MQA, GQA, MLA, FlashAttention — the complete mechanism that makes transformers work.
• The Transformer Architecture · 16 lessonsFrom 'Attention Is All You Need' to GPT-4 and Llama 3 — every layer, every design choice, every scaling decision.

Build

Turning a model into an application that does something useful.

• Building RAG Systems · 23 lessonsGround language models in your own knowledge — retrieval, reranking, evaluation, and production, end to end.
• Prompt Engineering and Context Design · 12 lessonsSystem prompts, structured outputs, chain-of-thought, and the context engineering patterns that make LLMs reliable in production.
• Vector Databases, Inside and Out · 17 lessonsAlgorithms, architecture, and operations — from HNSW internals to production at scale.
• AI Agents in Production · 16 lessonsFrom the agent loop to multi-agent systems — LangGraph, MCP, tool calling, memory, and the guardrails that make agents safe to ship.

Ship

Everything between a working prototype and a system you can operate.

• LLM Evaluation and Testing · 14 lessonsGolden datasets, LLM-as-a-judge, regression testing, and CI/CD gates — the engineering discipline that separates demos from products.
• LLM Fine-Tuning for Production · 14 lessonsLoRA, QLoRA, dataset engineering, and the decision framework for when fine-tuning is the right answer — and when it isn't.
• LLMOps: Serving and Operating LLMs · 15 lessonsModel serving, inference optimization, guardrails, observability, and cost control — the engineering that keeps LLM systems running.

Evolve

The patterns arriving now — agent harnesses, self-improvement, compound systems.

• Loop Engineering and Agent Harnesses · 14 lessonsHarness design, context management, subagent orchestration, evaluation-as-steering, and the Claude Agent SDK — the infrastructure layer that makes agents reliable.
• Autonomous and Headless Agents · 14 lessonsDurable execution with Temporal, event-driven triggers, sandbox isolation, crash recovery, and the governance framework for agents that run without human oversight.
• Self-Improving AI Systems · 13 lessonsAutomated prompt optimization, log-driven harness patching, principle extraction, memory-driven evolution, and the governance framework for systems that modify themselves.
• Compound AI Systems · 13 lessonsMulti-model routing, cascading, the classifier-generator-verifier triad, AI gateways, protocol composition, and the systems engineering that turns models into products.