Natural-Language Game Data Dashboard Agent

You are a decision-intelligence engineer building a natural-language data dashboard agent for a game studio: Users: [INSERT - producers, LiveOps, marketing, executives] Data sources: [INSERT - telemetry warehouse, revenue, support, store, A/B platform] Metrics catalog: [INSERT - DAU, retention, ARPDAU, funnels, cohorts] Access model: [INSERT - SSO, row-level permissions] Cover the full pipeline: 1. Design - question types the agent must answer, metric dictionary and definitions, decision workflows, when to route to an analyst. 2. Semantic layer - governed metrics, dimensions and time grain, ownership and certification, mapping business terms to warehouse tables. 3. Agent core - NL-to-query (SQL/API), query validation and dry runs, chart type selection, result explanation, multi-turn drill-down and comparison. 4. Intelligence - anomaly detection, automated attribution (segment, release, event, experiment), forecasting, scheduled briefings and alert digests. 5. Technology - warehouse connectivity, query governance and limits, caching, read-replica safety, audit logs, Slack/web/mobile clients. 6. Data quality and testing - golden question set, metric-accuracy tests vs known dashboards, hallucinated-number prevention, permission tests, load tests. 7. Launch - metric certification sprint, pilot team, feedback controls, analyst review of high-stakes answers, usage monitoring.