Design Mobile Model Usage Quotas

Scenario

You are an Android engineer building the model usage tiering and quota system for a mobile AI assistant app (e.g., ChatGPT). The system limits users on advanced models (e.g., 40 messages every 3 hours for Pro users, or 10 requests per day on Free tiers). When a user nears or hits their limit, the UI must react smoothly—displaying remaining allowances, countdown timers to quota reset, and upgrade or wait prompts without degrading message latency.

Requirements

Functional Requirements

• Client UX & State: Surface remaining quota and an accurate countdown timer for quota reset. Disable or adapt the send action when the limit is reached.
• Error & Edge Handling: Handle mid-stream quota exhaustion, rapid consecutive submissions, and out-of-sync multi-device usage.
• Atomic & Consistent Enforcement: Ensure the server remains the authoritative source of truth, while allowing the client to provide an optimistic, responsive UI.

Non-Functional Requirements

• Latency: Quota checking should not add round-trip overhead before dispatching a message.
• Reliability: Gracefully handle clock drift on the client and transient network failures.
• Scale: 50 million daily active users (DAU) with peak loads of up to 100,000 requests per second across all platforms.

What to Cover

1. API Contract & Synchronization Strategy: How quota status is transferred (e.g., dedicated polling vs. response headers in streaming SSE) and how the client tracks reset timestamps.
2. Android Architecture: Client-side state machine (e.g., MVI/MVVM), persistence layer (Room/DataStore), and integration with Kotlin Coroutines/StateFlow across multiple UI components.
3. Concurrency & Edge Cases: Race conditions when rapidly sending messages, speculative UI updates vs. rollbacks on 429s, and handling server-client time skew.