Video unavailable. Open the source below.
NewReceptionist Outcome Forecasting
Jev forecasts booking outcomes from structured receptionist-call traces and answers five post-call questions.
Added to Jevfast
How it uses Jev
The experiment sends structured receptionist-call traces to Jev, using turns, tool calls, workflow stages, and timing without transcripts or audio. After each agent turn it asks whether the call will end in a booking, and it also asks five typed post-call questions. The creator reports 38,012 turn-level forecasts, 10,145 post-call answers across 2,029 calls, 118 ms median latency, and evaluation against EHR booking outcomes; the reported AUC and ranking figures are creator claims, not independently verified. The question primitive and request fields are not disclosed.
Why it is interesting
It tests whether structured interaction traces can forecast a real downstream outcome without sending call transcripts or audio, and exposes both per-call traces and aggregate forecast curves for inspection.