Lesson 3: Reliability, what happens when the model does not answer

Module 7 · Lesson 3 · About 40 minutes · Needs: Lesson 2 (the answer engine, LangChain runnables); Module 1 Lesson 5 (backoff, idempotency keys)

The situation

Dev asks in the planning meeting: "What does the assistant do when the model provider is down?" Nobody knows. The answer engine makes one call and waits. If the call hangs, the customer watches a spinner; if it errors, the customer sees the error.

Priya adds a second problem. Aisha Rahman asked "When will my refund show up on my card?" and got two different replies, because her first request timed out and the page sent it again. Marta paid for both. This lesson wraps the engine so neither happens again.