The problem
Mortgage sales managers can only spot-check a handful of loan-officer calls, so most conversations are never scored: coaching lands days late, deals stall at the same stage for reasons nobody can see, and a compliance slip only surfaces after it's a problem.
Lending is heavily regulated — the wrong disclosure, an unqualified rate quote, or a fair-lending misstep on a recorded line is real exposure — but manual QA can't cover the call volume where those risks actually live.
The approach
We pointed our contact-center quality engine at the mortgage sales funnel: every call is transcribed and classified through an 8-stage funnel (from rapport and discovery to rate presentation and close), then scored against an 84-point diagnostic taxonomy spanning sales methodology, communication, objection handling, and compliance.
Regulatory breaches are treated as auto-fail flags that override the score, so a call that mishandles a disclosure can't grade out as 'strong.' Every score is tied back to the specific moment in the transcript that drove it, so a manager can audit the AI's read instead of trusting a black-box number.
The product ships as a live, self-serve demo: watch pre-scored sample calls (a strong Stage-7/8 close, a mediocre call that stalls at Stage 4, and a compliance auto-fail), or upload your own call audio and get it classified and diagnosed end to end.
What I built
- Real-time mortgage-call classifier with an 8-stage sales funnel and an 84-point diagnostic taxonomy across sales process, communication, product knowledge, objection handling, customer experience, and compliance
- Compliance auto-fail flags that override the composite score when a regulatory breach is detected, with the triggering transcript moment attached
- Stage-outcome scoring that shows exactly where a call stalled in the funnel and why, per diagnostic category
- A live self-serve demo: pre-scored sample calls across strong / mediocre / compliance-failure outcomes, plus upload-your-own-audio analysis
- Transcript-grounded rationale for every score so managers can calibrate and audit the AI rather than trust an opaque number
- TypeScript
- Next.js 16
- React 19
- tRPC
- Prisma
- PostgreSQL
- Google Gemini (Flash + Pro)
- Vapi telephony
- Server-Sent Events
- Railway
Result
A live, self-serve product that scores mortgage sales calls at volume instead of by sample — funnel stage, 84-point diagnostics, and compliance auto-fails — with every score traceable to the moment in the call that caused it.
- Publicly playable: watch pre-scored sample calls or upload your own for live analysis
- 8-stage funnel classification with an 84-point diagnostic taxonomy
- Compliance breaches auto-fail the call and are tied to the exact transcript moment