Modern LLM engineering

What you'll learn

Post-training · 7 lessons, about 5 hours

1. Base models vs assistants FreeWhat post-training adds to a pretrained model.
2. Supervised fine-tuningTeach behavior from example conversations.
3. Instruction and preference dataThe datasets that shape assistant behavior.
4. RLHFReward models and reinforcement learning from human feedback.
5. DPO and direct preference methodsPreference training without a separate reward model.
6. Constitutional AITraining with AI feedback guided by written principles.
7. Training reasoning modelsReinforcement learning on verifiable problems.

Adapting models · 4 lessons, about 3 hours

1. Prompting vs RAG vs fine-tuningPick the cheapest approach that works.
2. LoRAFine-tune a big model by training small low-rank updates.
3. Fine-tune an open modelRun a real fine-tune and evaluate the result.
4. DistillationTrain a small model to imitate a large one.

Inference · 4 lessons, about 2.5 hours

1. QuantizationStore weights in fewer bits to run models cheaply.
2. Serving modelsBatching, throughput and latency in production.
3. Speculative decodingDraft with a small model, verify with a big one.
4. Running open models locallyDownload, quantize and run a model on your own machine.

Systems & safety · 7 lessons, about 5 hours

1. Benchmarks and their limitsWhat leaderboards measure, and what they miss.
2. InterpretabilityLooking inside models to see what they compute.
3. Alignment and safetyMaking capable models behave as intended.
4. Multi-agent systemsCoordinating many agents on one large job.
5. Evaluating agentsMeasure multi-step, tool-using behavior.
6. Reading AI papersGet the idea out of a research paper in 30 minutes.
7. Capstone: a tuned, evaluated assistantFine-tune, evaluate and deploy a specialist assistant.

Before you start

Lessons here build on earlier phases:

• Phase 1: Building with LLMs
• Phase 2: Math, when you need it
• Phase 5: Transformers

The placement test can mark off what you already know, so you only study what's new to you.

What you'll build

A tuned, evaluated assistant

Fine-tune, evaluate and deploy a specialist assistant. Read the tutorial.

• Run a release review: train a candidate, evaluate it on every criterion, and apply release gates
• Diagnose a failed candidate and fix it by changing the training data
• Explain why a single headline metric, or a raw loss, can hide regressions

When you finish

Finish every lesson in this phase, or show you know it in the placement test, and you earn a certificate of completion, "Modern LLM engineering (Phase 6)", with its own link you can share.