Evals And Observability medium
After you swap to a smaller model, p95 latency falls from 900 ms to 350 ms and the ops dashboard is all green. Leadership asks whether answer quality is still fine. What is the right response?
- A Yes, because faster responses reduce user abandonment, and lower abandonment is itself the primary quality signal for a chat product.
- B Unknown until measured; latency and quality are independent, so run the offline eval and watch task completion before calling it a win.
- C Yes, provided error and timeout rates held steady, since availability plus latency fully characterizes the service quality users experience.
- D Compare output token counts between the models; if response length held steady, the smaller model is producing equivalent answers.
Answer: B. Unknown until measured; latency and quality are independent, so run the offline eval and watch task completion before calling it a win.
**Latency and quality are independent**: a smaller model can be faster and less accurate at the same time, so a cost and latency dashboard needs a quality metric beside it. Treating healthy ops metrics as a quality signal (C) is tempting because that is how non-LLM services are judged, but a model can return fast, error-free, wrong answers. The tradeoff is a latency and cost win against a quality loss that only evals or online task metrics will show.
One system design question a day, with spaced repetition so the ones you miss come back.
Get the appRelated questions
- In one release, a RAG assistant got a new embedding model and a new chunking strategy, and the end-to-end judge score dropped 6 points. The team only tracks the end-to-end score and has spent a week tweaking the generation prompt with no recovery. What is the best next step?
- Over two months, your team iterated through 40 prompt versions, each tuned against the same 150-case golden set. The offline score rose from 72 to 91. Online task completion and retry rate have not changed. What is the most likely problem, and what is the fix?
- A prompt change produced an impressive answer in a live demo for leadership. The PM wants to ship it today. What should you do?
- You plan to replace human graders with an LLM judge for your answer-quality eval. Before trusting its scores in a CI gate, what should you do first?
- Your CI gate fails any prompt change that drops the score 3 or more points on a 100-point rubric over 60 cases. Production generates at temperature 0.7. Running the eval twice on the same commit gives scores that differ by up to 4 points, and engineers have started rerunning failed gates until they pass. How do you fix the gate?
- Your RAG assistant's answer-quality score dropped from 82 to 70 after a change. You only measure final answer quality. Why is that a problem for debugging?