Eval Console · For the AI lead
The evals product, on the claims workflow
- The problem today
- Model choices for receipt reading are made on a few spot checks, with no record of why.
- What it does
- Dvar's Eval Console turns the same run into a workflow you can audit: traffic, an answer key, a rubric with must-pass checks, a judge you can calibrate, and release gates per model.
- What you see
- The console itself, read-only, on the test receipts of this demo's measured run. Pick a step to open that screen.
What to look at: Every receipt today's model read into a claim, with its market, reply time and cost. Open a row to see the trace; people's names are masked.
The Eval Console is built for a laptop screen. On a phone it opens below; scroll inside it, or open it full screen.
Open the console full screen ↗