30-day system design sprint · presale
You can read system design for months and still freeze at the whiteboard.
Reading gets you recognition. Interviews want production under pressure. So a diagram from the free HLD Gym book teaches itself, then it is taken away. You rebuild it from memory, explain it the way you would out loud, an interviewer picks holes in what you said, and you get an honest grade: what you missed, named, with your own words as the evidence. Six or seven of those a day for thirty days is 197 reps, one for every diagram in all 51 chapters. About an hour a day, set at the senior bar: Meta E5, Google L5 and equivalents.
A short form, then the payment link by email. You are charged nothing until you click it, and refunded in full any time before day one.
Not ready to buy? Read all 51 chapters free, no signup, or see everything the $19 buys.
try it right here
One real rep, running on this page
Ninety seconds. It will probably go badly. That is the entire point of showing it to you before you pay.
how a rep works
Six steps, and the diagram always disappears.
Watch a diagram once, then rebuild it without it. Six or seven different diagrams a day, and the ones you fail come back.
01 Watch
The diagram builds itself, step by step, with the reason for each one.
02 Lock
It vanishes. No scrolling back, no peeking. This is the part that works.
03 Rebuild
You put it back from memory, in words: every step, in order, the way you would say it out loud. No drawing canvas. What fails at a whiteboard is the sequence, not the pen.
04 Defend
An interviewer attacks the answer. What breaks when this node dies? What is your write path at 50k QPS?
05 Score
Graded against the chapter, not vibes. What you missed is named out loud.
06 Return
What you failed comes back in days. What you nailed comes back in weeks.
why the diagram disappears
The part that feels worst is the part that works
Re-reading feels smooth. Smooth feels like knowing. That is the most expensive illusion in interview prep.
In 2006, Roediger and Karpicke gave two groups the same passage and the same total time. One group read it four times. The other read it once, then wrote down everything they could remember, three times, with no feedback and no looking back. Five minutes later the readers were ahead. A week later the readers could produce 40% of the passage and the recall group 61%.
Before that final test, everyone predicted their own score. The readers predicted they would win. Reading it four times had been easy, and easy feels like knowing.
That feeling is what the lock removes. While the diagram is on screen you cannot tell recognising it apart from being able to build it. Take it away and there is nothing left to feel fluent about. You either produce it or you do not.
the obvious question
Why not just ChatGPT?
For explanation, you should. Explanation is free now, which is why the whole book is free. What a chat cannot sell you is a grade you can trust.
A chat is tuned to be agreeable. Ask it to quiz you and it praises a wrong answer delivered confidently, which is worse than no practice at all: it manufactures the exact feeling of readiness the lock exists to destroy. People find this out in public, every hiring season: flawless AI-assisted practice, then a failed loop.
The sprint sells the other layer. The diagram comes off the screen before you answer. The grade comes from the chapter the diagram belongs to, not from whatever the model improvises that day, and what you missed is quoted back from your own words. The schedule remembers which of 197 diagrams you failed twelve days ago; a chat window forgets the syllabus by Tuesday. That layer is the product. It is the one layer a free model does not offer, because honesty is not what it is tuned for.
15 a day · 450 across the sprint
The revision you will actually do
Ten teach the day’s topic. Five bring back one you finished, because nobody re‑opens a chapter from three weeks ago and everybody watches one more.
Four are built and playing below. The daily fifteen start on day one.
now playing · consensus and fencing
A lock is a statement about the past.
Told as a hotel keycard.
Click the reel to pause. Scroll it for the next one.
Ten reels on the day’s topic, five from a topic you already finished, chosen for you and mixed into one feed. Forty-five seconds each, one idea, told as something ordinary going wrong: a hotel keycard that still opens the door, three flatmates and a moved dinner, a coffee counter in a rush. In the last few seconds the same picture is relabelled as your system. Fifteen a day is about eleven minutes, the part of the sprint you do on the train.
They go where the book is hardest. Consensus, quorum overlap, isolation levels, clock skew: the chapters people bounce off once and never open again. A reel does not make them easy by making them wrong. It takes the one idea the chapter turns on and shows it happening to something you already understand.
Not because anyone is a “visual learner”: that idea has no evidence behind it (Pashler, 2008). Because a reel is the picture the interview asks you to draw, built one piece at a time, short enough that you finish it.
the source material
The book is finished, and it is free
The sprint is not the book. The book is what the sprint grades you against, and the sprint covers all of it.
| part | chapters | covers |
|---|---|---|
| The Meta-Game | 2 | how the interview is actually scored; company playbooks for Meta, Google, Palantir, Anthropic |
| Foundations | 14 | the request path, API design, data modelling and indexes, estimation, load balancing, caching, CDNs and blobs, replication, sharding and consistent hashing, CAP and quorums, queues, rate limiting, NoSQL, IDs, search indexes and locks |
| Senior Depth | 11 | 2PC, sagas and outbox; idempotency and ledgers; consensus, leader election and fencing; CRDTs vs OT; multi-region and cells; hot partitions; backpressure and circuit breakers; SLOs and error budgets; stream processing; probabilistic structures and gossip; zero-downtime migrations |
| The Problem Gauntlet | 24 | URL shortener, typeahead, notifications, WhatsApp, news feed, Uber, Dropbox, Ticketmaster, web crawler, global rate limiter, job scheduler, top-k, ad click aggregation, payments, Google Docs, distributed cache, Kafka, metrics, search, S3, YouTube, stock exchange, LLM inference serving, vector search and RAG |
Written for the senior bar: Meta E5, Google L5 and equivalents. No signup, no email, no paywall, and no plans for one. Open any chapter and judge the sprint by it.
It opens on what actually happened
36 of the 51 chapters start with the real incident that forced the mechanism into its shape. The other fifteen open on a scene.
Named people, named companies, a date, and a source you can go and read. The mechanism arrives as somebody’s fix for a bad night, which is what it was. Then the chapter teaches it properly.
What is in there
Counted off the built file, not promised. Open it and check.
- 36 histories, 189 sources. Papers, RFCs, postmortems, commits and oral histories, cited inline. Every date and figure is checked against the document, not a search result.
- 26 photographs and 46 company logos, licensed and credited, sitting in the chapter next to the paragraph that names them.
- 150 cards, four suits: person, incident, artifact, ripple. You keep a card by answering it, not by reading it, and the ones you keep join the same review queue as the chapter’s questions.
the offer
30-day system design sprint
Everything you are buying is listed below, and so is everything you are not.
Thirty days, one topic a day, the whole 51-chapter book.
- 6 or 7 reps a day, 197 across the sprint: a first sight of every diagram in the book, and what you failed comes back on top of that. Each is the loop on this page: watch, lock, rebuild from memory in words, defend, score, return. About an hour.
- An interviewer that pushes back: three follow-ups per rep, generated against your actual answer and aimed at what you left out.
- 15 reels a day, 450 in total: ten on that day’s topic, five on one you already finished. About eleven minutes.
- A schedule you never manage: what you could not rebuild comes back in days, what you nailed comes back in weeks. You open the day’s reps; you never pick them.
- Lifetime access: one payment, no renewal, nothing to cancel. It runs in the browser, on a phone or a laptop.
What happens after you click
There is no checkout on this page. The button opens a form, and payment is a link I send you.
01 You answer five questions
Sixty seconds: when you interview, at what level, where it breaks for you, which part of this you want most, whether you would pay today, and your email. No account, no password, no card.
02 I email you the payment link
Within 24 hours, to the address you gave. $19, one time. If you change your mind you simply do not click it.
03 You pay, and you are in the first cohort
Topic 1, its reps and its fifteen reels, then a new topic every day for thirty days. Nothing more is ever charged.
What you are not buying
Said plainly, so nobody pays $19 for a thing they thought was included.
- Not the book. All 51 chapters are free, now and permanently, with no signup. The sprint is the practice; the book is what it grades you against.
- No videos of anyone talking, no live calls, no Discord, no one-to-one mentoring, no certificate.
- No mock interview with a human. The interviewer is software; it grades you against the chapter, not against anyone’s mood.
- Not a subscription, and not a job guarantee. The book rebuilt from memory in thirty days, and an honest score, is the entire promise.
Before you pay
Answered by the person building it, before you spend anything.
- What exists today? The book, all 51 chapters, with 36 histories. The rep on this page, hand-built: its follow-ups were written in advance and its score is keyword matching. Model grading is built and running in a diagnostic interview that ends in a failure map quoting your own transcript; it is not hosted on this page yet, so everything you can click here today is the keyword version. Four reels of the 450. The sprint itself is not built yet, and reservation holders go first when it is.
- Why not quiz myself with a chat window? You should, tonight, for one rep. The difference is answered in its own section above: a chat will not take the diagram away first, will not grade you against a fixed chapter, and will not remember what you failed.
- My interview is in the middle of September. The form asks when it is. Say so, and the reply that carries the payment link tells you what the first days cover, before you have paid anything.
- What if I fall behind? Nothing expires. Access is for life, so the thirty days are the pace, not a deadline.
- What if it never ships? Every dollar back, without emailing me to ask. The same is true any time before your day one, for any reason.
- Why trust the sprint? Judge it by the book, which is by the same person. Open Consensus, Leader Election and Fencing or Ticketmaster: Booking Under Pressure and read at the depth the reps are set to. If the chapter is below the bar you need, keep your $19.
- Anything else? Email shreyashlrn@gmail.com or message me on LinkedIn. A real person reads these, and it is the same person building it.