Build an AI-Assisted Chat Backend

Using an AI coding assistant such as Claude or Codex, build a Python backend-only chat system from an empty project. No UI is required. First write a brief implementation plan, then implement the system while clarifying ambiguous requirements with the interviewer. The system must support at least: Sending and receiving messages A user can send a message to another user. Online users can receive messages addressed to them. Offline messages Messages sent while a recipient is offline must not be lost. When the recipient comes back online, they can retrieve messages received while offline. Conversation history A user can retrieve the conversation log with another user. Messages should be returned in a reasonable chronological order. Storage evolution Implement the core functionality with in-memory storage first. Then replace or abstract the storage layer for database-backed persistence with minimal impact on business-facing interfaces. After the core implementation, explain and discuss: Should delivery use polling, long polling, WebSockets/push, or a combination? Why? How should unread messages be defined and maintained? How do you handle persistence, duplicate delivery, retries, and retrieval after a user reconnects? How would you design APIs, data models, and a storage abstraction to migrate smoothly from memory to a database? You may define reasonable REST APIs, data models, and simplified authentication, but clearly state your assumptions and trade-offs.