14Your VP asks whether we should use an open-weight model instead of a commercial API. Lay out the trade-offs.▼hardNewMistral AIMetaDatabricks◆ premiumA trade-off question where both sides have serious arguments. The interviewer wants the dimensions that decide it, and a recommendation that depends on facts you would go and get.Practise this question →Read answer →
06Engineering proposes splitting our assistant into five specialised agents that talk to each other. Do you agree?▼hardNewMicrosoftGoogleSalesforce○ sign inMulti-agent diagrams look sophisticated in a design review. The interviewer is checking whether you can name the specific conditions that justify them and the costs they add.Practise this question →Read answer →
19Users say our assistant refuses harmless requests; legal fears it will answer harmful ones. How do you evaluate this?▼mediumNewAnthropicOpenAIMeta◆ premiumTwo teams pulling in opposite directions, each with a real point. The strong answer gives them one shared measurement so the argument becomes a decision.Practise this question →Read answer →
11Our AI search answers have a p95 latency of six seconds. Product wants under two. How do you get there?▼hardNewPerplexityGoogleFlipkart◆ premiumThe candidate who says 'use a faster model' in the first sentence has already lost points. Decompose the budget, find the biggest bar, and check whether two seconds is the right target at all.Practise this question →Read answer →
30The new forecast has four times the squared error, but operations wants it. Which model should ship?▼mediumNewAmazonUber◆ premiumTwo forecasts make errors in opposite directions. Calculate the operational consequences before deciding which evaluation metric should control the release.Practise this question →Read answer →
05Infrastructure proposes serving our model in 4-bit precision to cut costs by more than half. How do you decide?▼hardNewNVIDIAMetaMistral AIunlockedQuantization offers large savings with small average quality loss, and the averages hide where the loss lands. The interviewer wants evaluation on real workloads, by segment, before any cost win is claimed.Practise this question →Read answer →
15GPU capacity is short for the next two quarters. Research, the API business and two consumer products all want more. How do you allocate it?▼hardNewGoogleMetaMicrosoft◆ premiumA resource allocation question with no painless answer. The interviewer wants a transparent framework, efficiency gains before rationing, and decisions leadership can stand behind.Practise this question →Read answer →
23Write a launch recommendation: the AI draft feature saves time, but quality reviewers want a delay.▼hardNewAmazonAtlassianMicrosoft◆ premiumTurn conflicting pilot evidence into a recommendation an executive can approve. Show the decision, the rejected alternative, the workload cost and the evidence that would change your mind.Practise this question →Read answer →