Design an AI search engine.

### Signal to interviewer

I can design AI search systems that connect retrieval quality directly to user outcomes and trust.

### Clarify

I would clarify corpus type, update frequency, user intent distribution, and acceptable hallucination risk.

### Approach

Use a retrieval-augmented stack: intent parsing, hybrid retrieval, reranking, grounded synthesis, and citation checks.

### Metrics & instrumentation

Primary metric: task completion after search interaction. Secondary metrics: reformulation reduction, citation click utility, and freshness hit rate. Guardrails: unsupported claims, stale-source usage, and latency regressions.

### Tradeoffs

Higher recall increases coverage but can reduce precision and trust. Strong filtering improves reliability but may hide relevant long-tail results.

### Risks & mitigations

Risk: stale index undermines relevance; mitigate with incremental indexing. Risk: synthesis overclaims; mitigate with claim-to-source verification. Risk: retrieval bias toward popular documents; mitigate with diversity-aware reranking.

### Example

For enterprise policy search, the engine combines semantic retrieval with policy recency ranking and displays source timestamps in every answer.

### 90-second version

Build AI search as a grounded pipeline: understand intent, retrieve broadly, rank carefully, synthesize with citations, and validate claims. Optimize task completion while guarding trust.