🤖
AI Engineering & Agents
How language models work — and how to build with them.
From tokens and attention to retrieval, evals and autonomous agents. The durable concepts behind building reliable software on top of large language models, without the hype.
LLM Foundations
What a language model actually is, and why it behaves the way it does.
- Tokens and embeddings How LLMs chop text into subword tokens, why that shapes cost and odd failures, and how embeddings turn meaning into geometry. ●●○○○ 18 min
- Transformers and attention Next-token prediction, the query-key-value intuition behind self-attention, and why transformers displaced recurrent networks. ●●●○○ 22 min
- How LLMs are trained Pretraining, scaling laws, instruction tuning, preference learning and RL on verifiable tasks, and what "the model learned" really means. ●●●○○ 24 min
- Sampling, context and limits How LLMs pick each token, what temperature and top-p do, how context windows and cutoffs bound them, and why they hallucinate. ●●○○○ 20 min
AI Engineering
Prompts, retrieval, evals and the craft of shipping LLM features.
- Prompt engineering that actually works Clear instructions, context, examples and structure — and treating prompts as code you iterate against tests, not magic words. ●●○○○ 14 min
- Retrieval-augmented generation (RAG) Grounding a model in retrieved documents: chunking, embeddings, keyword and vector search, reranking, citations, and how it all fails. ●●●○○ 16 min
- Evals: testing non-deterministic systems Why evaluation is the core discipline of AI engineering, and how to build test sets, graders and regression checks you can trust. ●●●○○ 16 min
- Prompting, RAG or fine-tuning? A decision framework for adapting an LLM: prompt first, retrieve for missing knowledge, fine-tune for behaviour at scale. ●●●○○ 14 min
- Structured outputs & reliability Getting machine-readable output from LLMs, validating all of it, and designing retries and fallbacks so wrong outputs fail safely. ●●●○○ 14 min
- LLMs in production: cost, latency and guardrails Running LLM features for real: token economics, streaming, caching, model routing, rate limits, guardrails, drift and privacy. ●●●○○ 15 min
Agentic AI
Models that use tools, take actions and work in loops.
- What is an AI agent? The agent loop, how agents differ from predefined workflows, and when an agent is worth its extra cost and risk. ●●○○○ 20 min
- Tool use and function calling How models request tool calls from definitions, why your code runs them, and how to design tools an agent can use well. ●●●○○ 22 min
- The Model Context Protocol (MCP) An open protocol for plugging tools and data into AI apps: hosts, clients, servers, core primitives, transports and security. ●●●○○ 22 min
- Agent design patterns The core building blocks of LLM systems, from chains and routers to agent loops, and why the simplest pattern that works usually wins. ●●●○○ 20 min
- Agent memory and context engineering How agents manage a finite context window: curating inputs, compaction, retrieval, external notes and sub-agents. ●●●○○ 20 min
- Agent safety and evaluation Why agents that take actions raise the stakes, how prompt injection works, and how to test agents for real reliability. ●●●●○ 25 min