RingBack, behind the curtain

Built in about 3 hours. Here's how.

RingBack is a working product: real phone calls, researched pitches, a CRM that writes itself, automated follow-up email and calendar invites, and a signup funnel. One person built it, with AI agents doing the engineering and a human making the product calls. That split is the point.

The build timeline
Spec first

Deep research across voice-AI platforms (ElevenLabs, Retell, Vapi, Bland, DIY Twilio), verified against live docs, before a line of code. Decision criteria: time-to-first-ring, webhook quality, cost per minute.

Mock-first build

The entire product shipped and was testable before any telephony existed: simulated call lifecycle, same completion path as live calls. The demo worked end to end with zero API keys.

The pivot

The original carrier blocked number purchases on a compliance review the night before launch. The whole voice layer switched providers in about 90 minutes without touching the product, because the provider seam was designed in from the start.

Live and iterating

First real calls surfaced real bugs: webhooks eaten by a CDN, generic discovery questions on consumer products, a hallucinated claim in a follow-up email. Each got a structural fix the same night: self-healing status reconciliation, product-aware pain-funnel questions, factual grounding rules.

What happens on every call

Before the ring

  • Your homepage plus 1-2 key pages fetched and distilled into a sales brief: product, ICP, value props, brand tone
  • 3-4 pain-funnel discovery questions generated to fit YOUR product
  • All inside a hard 8-second budget, the phone always rings fast

On the call

  • Sandler structure: bond, up-front contract, pain funnel, close
  • Riley digs one level on every answer: what is it costing you?
  • Grounding rules: no invented features or integrations, ever

After you hang up

  • A CRM record whose fields fit your business, built from the transcript
  • Follow-up email and a real calendar invite, actually sent
  • Fit score with rationale, full transcript, call recording

The stack

  • Next.js + SQLite on Railway; Retell for voice; Claude for research, analysis, and writing
  • Runs for roughly $25/month plus about 13 cents per call minute
  • TCPA consent logging, rate limits, spend caps, voicemail handling
Why it exists

Built by Jim Breese as a working answer to a question every growth team is asking: what does one operator plus agents actually ship? The engineering ran through an agentic workflow; the product decisions, sales methodology, and taste are human. Try the call, then imagine this pointed at your funnel.