Voice AI receptionist build for HVAC business (n8n + Vapi)

I’m launching a small HVAC service business and want to test an AI phone receptionist before we open. I’ve been watching a few YouTube builds (Vapi + n8n) and I think I know roughly what I want, but I’d rather pay someone who’s actually built one to do it properly than fumble through it myself. What I need it to do on inbound calls: • Answer with a greeting and ask for the caller’s email • Look up the caller in a CRM (I’m okay using Airtable) • If new caller, collect name, phone, address and create a record • Book appointments to a Google Calendar (with availability check) • Reschedule or cancel existing appointments • Answer basic questions about pricing, hours, service area from a knowledge base doc I’ll provide • Hand off to a human number if the caller asks • Log every call (summary + outcome) to a separate Airtable table Tech stack I want: • Vapi for the voice front-end • n8n (self-hosted is fine, I have one) with an MCP server connecting Vapi to the workflows • Airtable for CRM, appointments, call logs • Google Calendar for the actual schedule • GPT-4.1 or whatever you recommend for the LLM I have a doc with our pricing, hours, services that I’ll send for the knowledge base. Things I care about: 1. Caller has to be told it’s an AI on the first turn (compliance thing). 2. Has to handle emergency calls properly — if someone says gas smell or no AC in summer it can’t just try to book them next Thursday. 3. Has to sound natural. I’ve heard a few of these and some of them are painful. 4. Needs to actually work end to end, not just a demo that breaks on the second call. Deliverables: • Working Vapi assistant • All n8n workflows imported into my instance (I’ll give you access) • Airtable base set up • A Loom or short doc walking me through how to update prompts/KB myself If you’ve built one of these before please share a recording or a workflow screenshot. I don’t need a fancy proposal, I just want to see proof you’ve done it. Also tell me one thing that usually goes wrong with these builds so I know you’ve actually shipped one.