You said Aurora's hardest problems are making live agent calls feel fast, simple and trustworthy. I work on a close cousin of that at Recruit41, so I gave it some thought.
The user waits while calls run. Answers should stream in, and they should be free to leave.
Thirty-six calls, one answer. Sum it up in a sentence, details on tap.
People are trusting an AI's account of a call. Every price should link to the moment it was said.
A working prototype. Click through the three steps.
Got it. Two quick things so the shops can give a real price:
"Hi, I'm an AI assistant calling for a customer in Indiranagar. Do you do jet service for a 1.5 ton split AC? What would it cost, and do you have a slot this Saturday?"