Section checkpoint

Ship to Production

8 questions drawn from the 4 chapters in this section. If most land on the first try, you are ready for the next section. If not, the explanation after each answer says which idea to revisit.

Checkpoint

8 questions · first-try answers earn the most XP

1. When should you evaluate whether the unit economics work?

2. Why does `max_tokens` matter for cost as well as for verbosity?

3. Why should retrieval run in parallel with prompt setup rather than after it?

4. Why does keeping vector search in the same region as generation matter?

5. What does the chapter say about a change that improves answers but doubles cost?

6. Why does the chapter recommend a weekly prompt and retrieval review?

7. What should you redact from prompts and responses before logging them?

8. Why does an agent's wrong answer have no obvious line number to inspect?

What this checkpoint covers

Understanding AI App Costs
Tokens, $/1M, caching to cut 80% cost.
Latency Optimization
Streaming, KV-cache, speculative decoding intuition.
The AI Engineering Lifecycle
Prototype → eval → ship → monitor → iterate.
LLM Observability & Tracing
Traces, spans, token/cost per call — debugging a chain you can't step through.

Next section: Ethics, Limits & Career