Sample mock interview report
What you get after a voice mock interview. Sample data, no account needed.
Voice mock · System design · 24 min
Senior AI Engineer at Northwind
Interview on Sun, Nov 29 · this was your 3rd mock for this job
How it went
Right building blocks; one trade-off left unexplained
Requirements first, then design, then risks
Only one number from your own work
Clear pace; a few fillers when pushed
Question by question
Question 1
Design a RAG assistant for a support team with 40,000 articles.
Content GoodStructure StrongEvidence Needs workWhat you said“I’d chunk the articles, embed them, put them in OpenSearch and retrieve the top five for the model.”
What workedYou started with requirements (freshness, latency, cost) before naming any tools.
What to fixYou never said how you’d know it works. Name the evaluation set and the metric you’d gate releases on.
Follow-up they asked “What happens when retrieval returns nothing useful?”
A stronger outline
- Requirements: freshness, latency budget, cost per question
- Ingest, chunk, embed; re-index when an article changes
- Retrieve, re-rank, answer with citations
- Evaluation set from real tickets gates every release (yours: 61% → 84%)
Question 2
You chose retrieval over fine-tuning. Why?
Content StrongStructure GoodEvidence GoodWhat you said“Fine-tuning is expensive and the articles change every week, so retrieval keeps answers fresh.”
What workedClear reason tied to the data changing, said in one breath.
What to fixAdd when you would fine-tune after all (tone, format, a small fixed domain). Showing the other side is what the follow-ups test.
Follow-up they asked “So when would fine-tuning be the right call?”
A stronger outline
- Articles change weekly, so retrieval stays current without retraining
- Fine-tune when the problem is style or format, not knowledge
- Could combine: a small tuned model for format, retrieval for facts
Question 3
How would you stop the assistant leaking customer PII?
Content Needs workStructure Needs workEvidence Needs workWhat you said“Um, we’d probably add a guardrail, and, like, filter the output.”
What workedYou knew guardrails belong on both input and output.
What to fixToo vague under pressure. Name where PII is removed (at ingestion), what checks the output, and who is alerted.
Follow-up they asked “What if the PII is in the article itself?”
A stronger outline
- Strip PII at ingestion so it never reaches the index
- Output check before the answer is shown; block and log on a match
- Alert an owner; review blocked answers weekly
Delivery
- 142words a minute (comfortable)
- 6filler words, 5 of them in question 3
- 1:48average answer length
Since your last try
42% 61%
Criteria met across the same questions. Structure improved most; evidence is still the gap.
Added to your plan
- TodayRe-answer “PII in prompts” out loud (5 min)
- MonTrade-off drill: retrieval vs fine-tuning, with push-back
- WedAdd one number to each of your three main stories
Weak answers come back in your plan before the interview, so you practice them again rather than forget them.
How the report is made: from the transcript of your spoken answers, scored against what this job asks for. Feedback quotes your own words. It is AI feedback, so treat it as a strong second opinion, not a verdict.
Get this report for your own interview
Add the job and its date. Free to start, no card needed.