Replacing airline IVR menus with the Jev decision model
A developer uses Jev for intent routing, dispatching a one-sentence request to the right team in 368ms at 0.97 confidence, scoring caller anger, and asking a clarifying question when confidence is low.
Every time I call an airline, I multitask on something else while the IVR menu reads out its options. And every time I miss the option I wanted and have to listen through them all again. So I built a simulation to replace IVRs with @typesafeai’s Jev decision model and using @cartesia for the voice agent. For example: a caller says one sentence (“my bag didn’t come off the carousel in Denver”) and Jev picks the right team to route it to in 368ms at 0.97 confidence. The IVR path for the same caller is 18 to 30 seconds of listening (assuming I’m not multitasking!). The same request also scores how upset the caller is, so an angry caller goes straight to a human. (Try doing that on IVR today.) A caller with vague needs, below a 0.95 confidence threshold, gets a clarifying question instead of a guess. I ran the routing demo in a command line. In production, @cartesia Ink-2 would transcribe the speech and Sonic 3.6 would speak the reply. You can also catch the full video on the @Cartesia developer channel. Link in the comments.