Practical Voice AI, built end to end
Patrick Michael is an independent Voice AI engineer based in South Africa, working remotely with clients. He owns the path from understanding the call problem and designing the conversation through integrations, testing, deployment, and operational follow-through.
The focus is the full call path: answer clearly, gather the right information, qualify when needed, book or route the next step, and hand useful details to the people who continue the work.
Production experience
- Five client agents live: took failing and incomplete Vapi systems through testing and deployment for a UK Voice AI company
- Onboarding reduced from days to about eight minutes: automated the path from a client's form and website information to a test-ready agent across five onboarded agents
- Calls connected to real operations: built message-taking, qualification, notification, and booking workflows using tools such as Make, Airtable, Twilio, GoHighLevel, Cal.com, and Google Calendar
- Complex calls made manageable: designed agents for inbound messages, sales conversations, booking, withheld-number handling, and cleaner call endings
How the work is done
- Start with the real call: what the caller needs, what the agent must establish, and what should happen next
- Build the whole workflow: conversation design, knowledge, tools, integrations, notifications, booking, and handoff
- Test in small, controlled steps: use call evidence and repeatable checks to find failures before treating the system as ready
- Keep a person accountable: Patrick owns the requirements, trade-offs, testing, and acceptance of the work
Experience behind the work
Patrick has worked across web development, systems analysis, client operations, digital marketing, and software delivery since 2001. That broader experience matters because a Voice AI agent is rarely useful on its own. It must fit the website, calendar, CRM, notifications, and people already handling the work.
Inspect the work
Try the public Voice AI demos, then visit Miles Dry Dock to see how call problems are investigated and improved. The aim is visible evidence, not a claim that every call is perfect.