Voice AI and call analytics for a healthcare founder

Context

A healthcare founder saw clinics drowning in phone volume — appointment scheduling, prescription questions, follow-up calls — and wanted an AI voice agent that could handle routine calls while surfacing analytics on every conversation. The deadline was investor-driven: a working product, not a deck, in weeks.

The challenge

Voice AI is unforgiving. Latency above a second feels broken, interruptions have to be handled mid-sentence, and in healthcare every transcript is sensitive data with compliance obligations attached. The MVP had to feel production-grade on day one — a demo that stutters in front of clinic buyers is worse than no demo.

What we built

• A real-time voice agent — telephony integration, streaming speech-to-text and text-to-speech, and an LLM conversation layer tuned for interruption handling and graceful handoff to humans.
• Call analytics — live transcription, intent tagging, and a dashboard showing call outcomes, durations, and escalation reasons per clinic.
• Compliance-first data handling — encrypted transcripts, retention policies, audit logging, and PII redaction in the analytics pipeline from the first commit.
• A demo mode — a rehearsable, deterministic call flow for investor and clinic presentations, because live AI demos deserve a safety net.

How it went

The founder went from signed scope to a product taking live calls within the MVP window, and used it to run clinic pilots and investor conversations in the same quarter. The analytics dashboard — originally a nice-to-have — became the wedge feature clinics asked about first.

Why it's representative

Zero-to-MVP engagements look like this: a non-technical founder, a hard deadline, a product category (voice AI) where cutting corners shows instantly — and a build scoped to prove the thesis without mortgaging the architecture.