ChatGPT-Style Chat at Scale
A system design round on a consumer chat assistant. Forty-five minutes, no code. Expect the conversations, the traffic and the budget to move.
The questions in this round
Design a ChatGPT-style chat product: millions of daily users, multi-turn conversations, answers streamed as they are written. Take me from a sent message to the first token on screen.
Power users keep one conversation going for weeks. Its history is now ten times the model’s context window.
A launch triples traffic overnight, and your main model provider starts returning rate-limit errors.
A user walks the assistant into a harmful answer over six turns, and a screenshot of it is circulating. Where does your design catch that?
Finance wants the cost of one conversation, and wants it down by a third without users noticing.