Community · E2 · artifact verified
Triage customer messages with seven parallel questions
A Portuguese customer-service demo where each client message makes one Jev call with seven typed questions in parallel - intent, urgency, sentiment and more - and a cost table prices it against routing everything to a large LLM instead.
01 · Role in the system
What Jev does here
Every incoming customer message triggers a single call carrying seven typed questions composed through the official typesafe_sdk with retry policy - the README tabulates each question, its type and what it is for, covering the classifications a service desk needs before routing. Answers come back calibrated in roughly 100-500 milliseconds, and the demo prices the run honestly: $0.042 per million input tokens, output free, with a baseline route showing what sending every message to a large LLM would cost instead. A recorded-replay mode stands in for the live API when no key is present, so the whole demo runs offline, and a presentation deck documents the comparison.
02 · Control boundary
Where Jev sits
Jev as the front-door triage of customer service: one call, seven parallel typed questions through the official SDK, calibrated answers route the message, and the demo prices Jev against the large-LLM baseline in the UI.
Code owns the loop, permissions, thresholds, validation, and side effects. Jev owns only the bounded judgments described above.
03 · Known limits
What this evidence does not prove
- Demo scope: Portuguese service flows, fictional messages, no production traffic.
- The recorded mode replays captured answers when no API key is set; live quality depends on jev-latest.
- Brand-new project (0 stars at review time).
04 · Attribution
Public sources
This is a Community record: the project was published by a third-party community author.
- fbadaro ↗Community · github · public · checked 2026-10-01