Cianplan

Level 6: Applied: cases, FDE scenarios, mocks · L6.3 AI-era cases

ChatGPT-like service

Design streaming chat, conversation storage and a GPU/gateway layer.

15 min · Pro

What you take away

  • 13M output tokens a second at peak needs roughly 13,000 accelerators, so the fleet is the system and the bill is about $38M a month.

And 3 more inside the class, in Pro.

How the class is laid out

  1. The prompt
  2. Requirements and the numbers
  3. The design
  4. Three decisions
  5. The twist
  6. Check your understanding