Terminal-style transcript of a call handled by Phil, the AI phone receptionist

AI Voice & Phone Agents

A CRM and an AI receptionist that answers a real phone line

MAC Septic

The problem

Mac Septic needed more than a scheduling app. Calls came in at all hours, jobs got missed between the field and the office, and there was no system tying dispatch, billing, and customer history together. Hiring a 24/7 answering service wasn't in the budget, and off-the-shelf CRMs didn't fit how a septic company actually operates.

What was built

  • Production CRM on FastAPI + Postgres + React 19, covering dispatch, billing, and customer history
  • 200+ API endpoints and 57 database migrations shipped incrementally without downtime
  • Phil, an AI phone receptionist on the Twilio → Deepgram → Claude → ElevenLabs voice stack (Pipecat), answering the real business line 24/7
  • Model selection driven by an eval harness scored against 51 real call transcripts — Haiku 4.5 beat Sonnet 98.1 to 95.8 while running about 30% faster and roughly a third of the cost
  • Integrations with Clover, Samsara, RingCentral, Twilio, Stripe, and Resend
  • Support for multi-state operations across Texas, Tennessee, and South Carolina

By the numbers

200+

API endpoints

57

DB migrations shipped

24/7

Call coverage

98.1 vs 95.8

Eval win margin

FastAPI PostgreSQL React 19 Twilio Deepgram Claude ElevenLabs Pipecat Stripe

Let's talk

Tell me about your business.

Fifteen minutes, no pitch deck. Tell me what's broken and I'll tell you honestly whether I can fix it.