ScaleMySystems lets you build a system from real components, push traffic, failures and attacks through it, and watch what breaks. You fix it, rerun the same traffic, and the numbers show whether the fix worked. It is for engineers preparing for system design interviews and for anyone who wants to reason about AI infrastructure before building it.
138 components mimic how the real thing behaves from its settings: latency distributions, connection pools, queues, CPU, quotas, crash rules, cache hit rates, replication lag and delivery guarantees. Nothing runs in a cloud.
The engine is a discrete-event simulation with a fixed clock and seeded randomness. The same design and seed always give the same result, which is how a fix is proven: same traffic, before and after.
352 problems, 133 of them on AI infrastructure. Each is a working system with a planted flaw, and an automated test proves it fails as given and passes once fixed.
The copilot and the AI review can only quote numbers that exist in your run; a validator rejects answers that invent one. Your chat messages are sent to Anthropic to answer them and are not stored by us.
It is a teaching simulator, not a load-testing tool or a cloud calculator. Behaviour follows queueing rules and the numbers you set, not measurements of one provider, and the cost estimate uses illustrative AWS list prices. Use it to understand bottlenecks and trade-offs, not to sign off a capacity plan.
No subscription. Five problems are free, each other problem is unlocked once with credits (everything inside it is then free), and new accounts get 100 free credits, and the price of every action shows before you spend. See the full price list.
Questions, bugs, refunds, privacy requests or ideas for problems: send a message and we reply by email.