Make each chapter a practice session.

Who this is for

You should be comfortable reading application code, designing APIs, using data stores, and debugging software. You can come from frontend, backend, mobile, or infrastructure work. The LLM track introduces the needed model concepts without requiring ML research experience.

Choose one track for focused study. Alternate the tracks when you want classic systems ideas to reinforce the AI applications. Use chapter outcomes and rubrics to skip material you can already explain.

Use the full learning loop

1. Read the outcome, mental model, and lessons. Trace the worked example with its assumptions visible.
2. Write a prediction before exploring a lab. Compare the result with your reasoning.
3. Produce the assignment artifact, change a constraint, and check it against the rubric. Then open the sample artifact to compare assumptions, reasoning, and explicit checks; revise the weak part.
4. Answer the recall questions without notes. Compare reasoning with the sample answer, then rate your practice.
5. Run the interview drill aloud, review the approach afterward, and repeat the weakest part.
6. Keep one cumulative project document per track: Relay for systems, Harbor for LLM engineering. Each chapter asks you to recover earlier ideas before adding a new decision. Preserve your first attempt and revision; the final chapters combine them into a complete design and recorded review. This work uses part of the assignment and recall budget rather than adding another course.

Paper and a text editor are sufficient. You can use fictional model outputs and workloads for the exercises. Paid APIs, cloud accounts, and external deployments are optional.

What the hours mean

The systems track budgets about 56 hours and the LLM track about 73 hours for a full practice pass. Each chapter shows a breakdown for reading, the assignment and variations, repeated drills, and recall. These are initial estimates for independent study. Split long assignments across sessions and move on when the rubric is satisfied.

Start with a pace you can repeat, such as two or three focused sessions a week. Logged practice rated 1–2 comes back for the next session; stronger practice returns after 48 hours. A missed knowledge check also prompts review. Completion is your judgment rather than an automatically graded credential.

A suggested sequence across both tracks

1. 01Build the machinery

   Systems chapters 1–4 with LLM chapters 1–3: requirements, storage, caches, generation, transformers, and training.

   Systems foundations →LLM foundations →

2. 02Add distributed behavior

   Systems chapters 5–7 with LLM chapters 4–10: queues, consistency, reliability, serving, tools, agents, retrieval, and memory.

   Distributed systems →Serving and applications →

3. 03Practice production judgment

   Systems chapters 8–12 with LLM chapters 11–18: operations, security, global systems, evaluations, product architecture, and infrastructure.

   Production systems →Production AI →

4. 04Combine and explain

   Systems chapters 13–15 with LLM chapters 19–21: complete designs, tradeoff records, timed mocks, and concise explanations.

   Systems practice →AI design practice →

Your progress stays in this browser

There is no account or network sync. Completion, quiz attempts, logged practice, bookmarks, and notes are stored locally. Export progress from the home page before changing browsers or devices, and import that file to continue. Clearing browser data can remove progress. Lab predictions and unsubmitted recall drafts last only for the current page.

Sources and updates

Every chapter links selected primary documentation or papers. Review dates describe the linked resources; current model names, prices, API support, and limits still need verification when you implement a system. Worked examples use explicitly illustrative values.

Practice material last updated October 2026. Course updates focus on clearer examples, corrected technical claims, and current official references. Retrieval practice and spaced study inform the study loop; see Dunlosky and colleagues’ review of learning techniques.