DEMO / RECORDING
Overview & result
An MIT-licensed open-source app uses Jev to coach a person during sales calls. The author’s follow-up says the demo responds in about a second and that the same pattern could be adapted for interviews or technical support.
From the original post
[ZeroGold/call-coach-ai: Jev powered call coach](https://github.com/ZeroGold/call-coach-ai)
Has an MIT license so you can just use it, rebuild it, do whatever!
How Jev fits in the loop
- Transcribe a live sales call.
- Use Jev to classify cues and score coaching guidance.
- Surface feedback to the caller in the interface.
How to reproduce
- Clone https://github.com/ZeroGold/call-coach-ai and follow its README.
- Run the call-coach demo with the required transcription and Jev configuration.
- Measure end-to-end timing and confirm consent, privacy and transcript handling before using real calls.
Why this build matters
A concrete realtime audio use case with a visible privacy boundary and a community-reported latency concern.
Limitations
- The source post does not publish a reproducible accuracy or latency benchmark.
- A commenter flags privacy and legal concerns when third-party call audio is sent to a remote service; the author says Jev cannot currently run locally and suggests a local model substitute.
- A community follow-up notes that overlapping speakers and tail latency need separate evaluation.