Voice AI interviews for high-volume hiring

The problem

Hiring for remote and outsourced support roles brings in more applications than a recruiting team can read, let alone call back. Strong candidates wait in the same queue as everyone else, and the first conversation, the one that shows how someone speaks and thinks, happens late or not at all.

What we built

We built the hiring flow end to end: a branded careers site, AI screening of every application against the role's own rubric with reasons a recruiter can check, and a four-step pipeline from first screen to final decision. A voice agent holds first interviews by phone or in the browser and returns each call as a scored transcript. The AI never asks about or discusses pay; those questions go to a person, and recruiters make every advance, hold or reject call.

What production made hard

• A voice interview has to feel like a conversation over an ordinary phone line, so delays and interruptions were tuned on real calls.
• A score helps only when a recruiter can see why, so every score points to the transcript lines behind it.
• Some questions should never come from a machine, so pay was routed to people by design rather than by instruction to the model.

What a similar project needs

• A rubric per role, written by the people who hire for it, before any model scores anyone.
• Consent and data handling for recorded interviews, settled before the first call.
• A review loop where recruiters correct scores, so the rubric improves with use.