SubstackGitHubBook a call →
← Back to side projects
Demo

Cosmo Labs

Building voice AI agent demos for real business workflows.

What I built

A set of voice AI agent demos built using Retell AI, designed to showcase what production-grade voice agents look and sound like in real business contexts — booking, lead qualification, and support triage.

Why it matters

Voice AI is moving fast but most demos are shallow — they show a single exchange in perfect conditions. These demos are designed to handle real edge cases: ambiguous answers, pushback, escalation requests, and fallback paths. The goal is to help buyers understand what they're actually getting before they commit.

How it works

  1. 1Each agent is configured in Retell AI with a structured script, intent detection, and escalation logic
  2. 2Agents handle common edge cases: unclear input, off-script questions, escalation to a human
  3. 3Call recordings and transcripts are reviewed to identify failure patterns and improve scripts
  4. 4Demo scenarios are designed around real SMB use cases, not synthetic examples

Demo scenarios

Booking & scheduling

Agent handles inbound appointment requests and calendar coordination.

Lead qualification

Agent qualifies inbound leads on key criteria and routes to the right rep.

Support triage

Agent handles tier-1 queries and escalates with full context attached.

Stack and tools

Retell AINext.jsTypeScriptVercel

Built with Retell AI voice agents to prototype and showcase agent behaviour in production-realistic scenarios.

What I'd improve next

  • Add call analytics to track drop-off points and intent signals per call
  • Test with real SMB customers to validate the booking and triage flows against actual usage
  • Add a CRM integration to log call outcomes and follow-up tasks automatically