Evals And Observability easy
Your team is starting an LLM feature that summarizes customer support tickets. A teammate suggests writing prompts first and building an eval set once the feature 'feels done'. What is the better plan?
- A Ship the prompt to a small beta group first, then build the eval set later from whatever user complaints come in over time.
- B Use a public summarization benchmark score as the quality bar until the feature is fully built and stable.
- C Collect 50 to 200 representative tickets with expected outcomes before writing prompts, and test every iteration on them.
- D Have the engineer who writes the prompt review a handful of outputs by eye after each prompt change is made.
Answer: C. Collect 50 to 200 representative tickets with expected outcomes before writing prompts, and test every iteration on them.
Building the eval set first costs some upfront labeling effort, but it gives a fixed target so every prompt change is measured, not judged by feel. Waiting for beta complaints is tempting because it feels like real data, but it is slow and biased toward loud failures. Complaints are a good way to grow an existing eval set later, not a substitute for having one.
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?
- 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?
- 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?