NEWAn independent referee for voice calls

A referee for every customer call — AI or human.

Voice Copilot sits alongside your calls and works to keep each one on track — watching for mistakes, checking the rules, and scoring every call afterward by the same standard. Independent, because no one should grade their own homework.

▶ Watch the demo
Live Call · 02:14
Customer: Jordan P.•Agent: Suraj
CUSTOMER "This $400 charge — I never authorized it!"
AGENT "That's frustrating, let me pull it up right now."
CUSTOMER "I want it removed today, not next week."
AGENT "Absolutely. I see it now — refunding to your card."
CUSTOMER "Okay… that's actually fast. Thank you."
CUSTOMER "This $400 charge — I never authorized it!"
AGENT "That's frustrating, let me pull it up right now."
CUSTOMER "I want it removed today, not next week."
AGENT "Absolutely. I see it now — refunding to your card."
CUSTOMER "Okay… that's actually fast. Thank you."
VOICE TWIN
528ms
⚡ Live Coach · de-escalation
P5
Acknowledge the bill shock first. Try: "That's a frustrating surprise — let me pull it up right now."
SENTIMENT · per turn
● agent ● customer
Customer: frustrated → calm+4 turns to resolve
How it works

A referee that watches every call — and steps in when it counts.

This is the heart of Voice Copilot, and it works hardest for AI voice agents.

During AI calls

Catches & corrects

An on-prem Nemotron model is wired into the AI agent's path, built to spot a wrong answer and correct it before it reaches the customer.

During · then after

Keeps it in bounds

During the call it checks permitted calling hours and timing. After the call, it reviews the full India RBI disclosure checklist — so nothing required is skipped.

After every call

Scores it the same

A post-call scorecard grades every call — human or AI — on one rubric. No one grading their own homework.

And a coach in the ear of every human agent.

Live coaching

A suggested response, surfaced while the conversation is still going.

Reads the room

Tracks how the customer is feeling, turn by turn, so tone can shift in time.

Clearer voice

Speaks the agent's words back in clean, easy-to-follow English on calls where accent differences get in the way.

Built to be checked

Not just claimed.

We built a repeatable test kit on NVIDIA GPUs — using NeMo Data Designer to generate tricky test calls and Nemotron models to score the results — so selected referee behavior can be tested reproducibly instead of taken on trust.

▶ Watch the referee score a call
FAQ

Questions ops leaders actually ask.