About this demo

I built Triage Desk to go with my article AI Decision Models Explained: Clef-Flash vs Jev. The article explains what decision models are, where they beat an LLM, and where they fail. This page lets you check those claims against a live model.

What is a decision model?

A decision model answers questions you define, using answer options you define. It returns a probability for each option and generates no text, so there is no JSON to parse and there are no output tokens to pay for. You set the questions per request, with no training step and no labelled dataset.

The Triage tab

Paste a support ticket or pick a preset. One call answers five questions: is it urgent, which team should handle it, how severe is it, does it contain personal data, and how upset is the customer. The threshold slider is the production pattern in miniature: at or above the threshold the ticket is routed automatically, below it a person decides.

The Break it tab

Five experiments show the limits. A choice question with no “other” option assigns even a pizza order to a team. The model estimates date arithmetic instead of calculating it. A vague question gives a number you cannot act on, rewording a question moves the probability, and the response never includes a reason.

What it costs

Clef-Flash is billed at $0.038 per million input tokens and output is free, so one decision here costs a few thousandths of a cent. The stats under each result show the latency, the token counts and the cost of that call.

Using a decision model in your own systems

I'm Thomas Wiegold. I build AI agents and automation for businesses, and classification steps like this one are a common place to cut LLM cost and latency. If you want to know whether a decision model fits your support inbox or document pipeline, see my AI consulting in Sydney, or find more of my work at thomas-wiegold.com.