1. Design it, interview-style
Requirements, rough estimates, API, a high-level diagram, the data model, then deep dives and trade-offs: the order you’d say them in an interview.
System Design Lab
System design interview prep, learned by building. Design each system the way an interview asks, then run the parts interviewers push on, so the trade-offs are something you've seen rather than memorised.
How each topic works
2. Build only the deep dives
Small Go labs for the pieces interviewers probe. Each step ends with something you run and watch: curl, logs, killing a process.
3. Finish with a cheat sheet
The short version to say out loud, plus answers to the follow-up questions you’re likely to get.
Run labs in your browser
Every lab page has an Open this lab button: a GitHub Codespace with Go, Claude Code and the ports ready. No local setup needed.
Learning path
Interview skillsBack-of-the-envelope estimates: load from user actions, what one server can take, how many servers that means.
1. Load balancerL4 vs L7, algorithms, health checks, 'isn't it a single point of failure?' Lab: build a round-robin reverse proxy.
2. Rate limiterWhere limits can live and what each layer can see, keying on client identity, sharing counters. Coming next: algorithms and a lab.
3. Chat systemWebSockets, routing messages across servers, L7 proxies vs direct connections, connections per server, multiplexing.
4. Distributed cacheConsistency per kind of data, one authoritative copy, leases and fencing, consistent hashing and hot keys, where routing lives.
Side topicsDNS, end-to-end encryption, the limits you hit when generating load, and choosing a language for systems work.
Learning with Claude Code
The repo’s CLAUDE.md sets Claude Code up as a tutor: one concept per step, the why before any code, a pause after each step so you can run it, and links back to interview questions. Open a lab in Codespaces, run claude in the terminal, and try:
- “Walk me through the rate limiter interview design, Part 1.”
- “Let’s build step 1 of the rate limiter.”
- “Quiz me on the load balancer cheat sheet.”
- “Design a URL shortener as a new experiment, same structure as the others.”
When you’re done, run /wrap-up. Claude saves a summary of what you learned to your own notes in the repo and drafts fixes wherever these lessons fell short for you. You choose what to send, and it goes in as a pull request for review. That’s how the lessons get better for the next reader.