Soundboard: Conversational Voice Analytics

Modern business intelligence is plagued by friction: executives and operators wait days for data teams to build reports, or struggle through complex SQL builders and rigid dropdown filters. Soundboard eliminates this bottleneck by making data exploration as immediate as a conversation. Powered by AssemblyAI's full-duplex streaming voice agent, Soundboard lets users ask questions out loud and watch live, responsive dashboards assemble on screen in real time. Rather than generating simple text answers, the system executes client-side tool calls that render interactive bar charts, time-series area trends, and category donut breakdowns in under 100 milliseconds. Reliability and enterprise safety are foundational. Most AI analytics platforms hallucinate raw SQL that fails or introduces security vulnerabilities. Soundboard bypasses raw SQL generation entirely. Spoken queries compile deterministically through an allowlisted YAML semantic layer with strict parameter binding, dimensional synonym resolution, and read-only database roles. Follow-up commands such as breakouts, period comparisons, and granularity changes modify existing visualizations in place rather than starting from scratch. Full-duplex barge-in allows users to interrupt the agent mid-sentence without lag or state corruption. When an analysis is complete, users can inspect the compiled SQL with full parameter transparency or export the entire dashboard directly to Metabase with a single click. Soundboard has been validated against 30 automated conversational benchmark scenarios covering speech ambiguities, multi-turn corrections, and edge cases, achieving a 100% pass rate with a median latency of under 45 milliseconds.