← All courses
Course 2

AI engineering, hands on

Build applications on top of large language models: how they work, prompts and context, retrieval, tools and agents, evaluation, safety and production. Same format: predict, watch it go wrong, fix it, compare.

11 hands-on lessons
Start: RAG: answering from your documents →
  1. 1 · PredictCommit to a guess first.
  2. 2 · WatchSee it break, slowed down at the key moment.
  3. 3 · FixTurn on one plain-language fix.
  4. 4 · CompareReal before and after numbers.

1 · How LLMs work

3 modules
Module 1Coming nextTokens and the context windowWhat a token is, what you pay for, context limits, truncation, long prompts.
Module 2Coming nextSampling and temperatureTemperature, top-p, repeatable answers, when creativity hurts.
Module 3Coming nextChoosing a modelBig vs small, open vs closed, cost, speed and quality, routing between models.

2 · Prompts and context

3 modules
Module 4Coming nextPrompt engineeringSystem prompts, examples (few-shot), step-by-step reasoning, clear instructions.
Module 5Coming nextStructured outputJSON the code can trust, schemas, validation and retries.
Module 6Coming nextContext engineeringWhat goes into the window, memory, summarizing long chats, context that rots.

3 · Retrieval (RAG)

2 modules · 11 lessons
Module 711 lessonsRAG: answering from your documentsSearch by meaning, chunk size, top-k, reranking, hybrid search, stale and private pages.
Module 8Coming nextVector databasesEmbeddings at scale, approximate search (HNSW), recall vs speed, filters.

4 · Tools and agents

3 modules
Module 9Coming nextTool callingLetting the model call your functions, arguments, errors, parallel calls.
Module 10Coming nextAgentsThe agent loop, planning, getting stuck, step and cost budgets, multi-agent.
Module 11Coming nextMCP: connecting toolsThe Model Context Protocol: servers, clients, tools and resources.

5 · Quality and safety

3 modules
Module 12Coming nextEvaluating AI systemsTest sets, exact checks, AI as a judge, regression tests for prompts.
Module 13Coming nextHallucinations and guardrailsChecking answers, refusing safely, input and output filters.
Module 14Coming nextPrompt injection and AI securityInjected instructions, data leaks, tool misuse, least privilege.

6 · Running it in production

3 modules
Module 15Coming nextCost and latencyPrompt caching, streaming, batching, small models first, rate limits.
Module 16Coming nextLLM observabilityTracing every step, user feedback, catching quality drift.
Module 17Coming nextFine-tuning or not?Prompting vs RAG vs fine-tuning, data quality, LoRA, when it pays off.

7 · AI engineering labs

1 module
Module 18Coming nextAI product labsSupport bot, document Q&A, coding agent: build each, then make it reliable.