JevMade hello@JevMade.com
← Back to guides

JevMade field notes / Evaluation guide

Grading automated support messages with a multiple-choice AI tool

This guide explains how to use Jev, an AI tool that chooses from options instead of writing answers, to grade automated support messages. It shows how to compare the software's grades against actual human reviews.

Original by Mohammed SanjeedEvaluation

Listen to this guide

JevMade’s plain-English explanation

0:00 /

AI narration

Credits

“How to Use TypeSafe AI’s Jev as an LLM Judge” by Mohammed Sanjeed. Read the original source.

This expanded guide is an AI-narrated adaptation prepared by JevMade. It expands the source’s essential ideas, examples and caveats in JevMade’s own words and is not a word-for-word reading. The synthetic voice does not imitate the author or imply their endorsement.

Our summary

Businesses use AI assistants to answer customer questions, but these programs can sometimes give unhelpful replies. The author tests Jev, an AI tool that chooses from options rather than writing an answer, to grade those replies for specific qualities like clarity and helpfulness.

You give the tool a customer conversation and ask it specific questions, like whether the software actually answered the user's request. The tool returns a percentage showing how likely the answer is yes. The author suggests comparing these scores to human grades to adjust the tool's accuracy.

This method helps teams managing automated support systems. The tool cannot do math, count, or write explanations. The initial test uses a single invented conversation, so checking the tool against real human reviews is a required step, not a guaranteed outcome of the process.

Key takeaways

  1. Ask the tool to check specific problems instead of asking for one overall quality grade.
  2. Inspect disagreements between the tool and human reviews, keeping untouched human examples for final testing.
  3. Lock in the specific software version you tested so unexpected updates do not change your results.

The tool cannot verify that a software assistant transferred a customer just because the message says so. You must include a real system record of the transfer for the tool to check it.

HoneyHive guide · Source reviewed · Original published

Read the original guide Opens the author’s site in a new tab.