I built CallSetter AI's voice-agent product end to end: prompt engineering, QA, and the CRM integration that made it hold up in a real contractor's workflow instead of just a demo. The company has since been sold. This is what I actually built, and the engineering that went into it.
The Problem We Were Solving
CallSetter AI was an inbound voice-receptionist product for US home-service contractors: HVAC, plumbing, pest control, landscaping, electrical, solar, cleaning. When a lead calls a small contracting business after hours or in the middle of a job, nobody picks up, and the lead calls the next name on the list. I built an AI agent to answer instead, qualify the lead on the spot, and book straight onto the calendar.
What I Built
My role was the engineering and the debugging: getting the product to actually work and hold up under real calls, not just controlled demos.
- Voice-agent prompt engineering and QA: built and refined agent prompts against a growing benchmark of real test calls, not just happy-path scripts.
- GoHighLevel CRM integration: the piece that would otherwise have needed a dedicated contractor developer to build.
- Per-niche demo pipeline: automated generation of "hear your agent" pages tailored to each industry vertical.
- A2P-compliant messaging infrastructure: Twilio A2P 10DLC approved, feeding a four-branch omni-channel follow-up flow across booked, abandoned, no-show, and showed leads.
Who It Was For
Home-service contractors are where this problem is sharpest: high call volume, thin office staff, and a caller who moves to the next name on Google the moment a call goes unanswered. That's why the product was built for HVAC, plumbing, pest control, landscaping, electrical, solar, and cleaning businesses specifically, where the cost of a missed call is immediate and visible, not abstract.
The Outcome
CallSetter AI has since been sold. That's the proof that matters most to me: real agents answering real calls, a CRM integration that held up, and messaging infrastructure that passed carrier compliance, engineered to a point where someone else was willing to take it on.
Common Questions
What was your role, exactly?
Engineer and debugger. I wrote and refined the voice-agent prompts against real call transcripts, built the GoHighLevel CRM integration, and got the messaging infrastructure through carrier compliance. If something broke on a live call, fixing it was mine to do.
What made it sellable?
Everything shipped and worked in production, not just in a demo: agents answered real calls, the CRM integration held up, and the compliance infrastructure passed carrier review. A buyer can trust a product that's already proven under real usage.
The Bottom Line
CallSetter AI shipped a real product and found a buyer. What I carry forward from it is the discipline of QA-ing an AI system against real inputs instead of shipping on vibes, and the ability to wire AI into a client's actual CRM and messaging stack rather than bolt it on top. That's the layer running underneath the paid-ads work I do now: campaigns that get optimized by more than gut feel.
Want that layer applied to your ad account?
Thirty minutes, no slides. Bring what you're running and we'll look at it together.
Book a call