AI Speech Analytics

Voiso is a CCaaS platform processing tens of millions of calls a month. Speech Analytics is its AI layer: transcription in 10+ languages, sentiment, topic detection, conversation scoring, and call summaries. Its home is Call Detail Records, the page where supervisors evaluate every conversation. When this project started, the AI pipeline could already produce the data, but the legacy page could not show it: the indicators sat scattered and mostly invisible.

Voiso made a strategic bet to step into AI analytics as the platform's competitive move. The CDR page was already queued for its transition to the new design system, and I framed the two as one project: rebuild the page on design system components and use that same pass to launch the killer feature, not a visual update with AI bolted on later.

I took the design fully hands-on and rebuilt the entire page on the new design system, extending it where AI needed primitives it did not yet have: waveforms, sentiment and topic markers, transcript patterns. Every screen was designed by hand across six discovery rounds, before AI design tooling existed. I worked in a tight four-person loop spanning product leadership, product management, and the LLM engineer, shaping what the models should return alongside how it renders. The structural calls were mine: a synchronized wave stack putting agent, customer, sentiment, and topics on one timeline; an event-marker system refined through four rounds of shape and state work and settled in a formal design review. The final component matrices, channel by persona by feature by view, served as the engineering specification: the development tasks were named after my frames.

• Transformed large volumes of call recordings into actionable operational insights.
• Reduced manual effort required to review conversations and identify quality issues.
• Enabled managers to monitor performance at scale.
• Increased visibility into customer interactions and team performance.