Design an AI copilot for data analysts.

### Signal to interviewer

I can design analyst AI that accelerates insight generation while preserving data correctness and trust.

### Clarify

I would clarify analyst persona, data stack, governance requirements, and common analysis workflows.

### Approach

Build a query-to-insight assembly: intent parsing, SQL generation with safety checks, result validation, and narrative/chart synthesis.

### Metrics & instrumentation

Primary metric: time-to-validated-insight. Secondary metrics: query correction rate, reusable analysis templates, and dashboard publishing velocity. Guardrails: semantic SQL errors, metric-definition drift, and unsupported narrative claims.

### Tradeoffs

More automation reduces manual effort but can increase hidden analytical mistakes. More transparency improves confidence but adds extra interaction steps.

### Risks & mitigations

Risk: wrong joins or filters; mitigate with schema-aware planners. Risk: hallucinated narrative claims; mitigate with source-linked statements. Risk: overreliance by junior analysts; mitigate with explainable reasoning panels.

### Example

For revenue analysis, AI proposes SQL for cohort retention, flags missing date filters, and generates annotated charts tied to validated result sets.

### 90-second version

Design analyst copilots for speed plus verifiability. Keep SQL and assumptions inspectable, enforce guardrails, and measure validated insight throughput as the core outcome.