AI Evaluation

Compare model and prompt quality across task success, groundedness, latency, cost, and failure categories before shipping changes. The screen connects Suites and Runs so platform and engineering teams can find the “Run eval” action while checking pass rate and cases.

Use this reference when you need to design this workflow inside an existing product. The example includes practical records such as Task success · 94% and Groundedness · 91%, rather than an empty visual shell.

AI Evaluation
Use this template in Figr

Template facts

Screen

Analytics

Primary action

Run eval

Pattern

Analytics

State

Default, loading, empty, error, permission, success

Design system

Adaptable product UI theme

What this screen helps users do

Platform and engineering teams need a clear “Run eval” action without losing sight of pass rate, cases and the affected records. Keep the decision and its outcome clear before moving to the next task.

Key elements

  • Primary action: Run eval is clearly available where the task happens.
  • Main view: Suites keeps the starting point obvious.
  • Connected view: Runs supplies nearby context.
  • Metric: Pass rate is visible with a meaningful value.
  • Another signal: Cases helps people judge what to do next.
  • Representative record: Task success · 94% makes the example specific.
  • Second record: Groundedness · 91% shows the pattern works beyond one item.

States worth designing

  • No data for the chosen period; show what people can do instead of leaving a blank panel.
  • Loading suites without changing the layout unexpectedly.
  • Partial or delayed metrics; explain the problem and preserve work.
  • Permission denied for runs; clarify who can continue.
  • Success after choosing Run eval; confirm the result and the next step.
  • Network or data refresh failure; offer a retry without losing context.

Where this screen sits in the flow

Suites → Runs → Run eval → Review outcome

Patterns used

Analytics, Suites, Runs, Cases, accessible states, contextual actions

Why these choices work

Make time ranges, filters and metric units clear. Show the data behind each trend so readers can test what changed.

In this design, people should be able to compare Task success · 94% with Groundedness · 91%. Make pass rate clear, then use cases to explain whether the next step needs attention.

Before handing this to a designer or engineer, check keyboard focus, error copy, permissions and what happens after they choose the “Run eval” action. The happy path alone is not enough.

Related UI examples: Prompt Library, Agent Runs, Model Playground.

Start from this in Figr

Design a ai evaluation screen for our existing software product. Use our real design-system tokens, components and product language. Show Suites and Runs, include believable records such as Task success · 94%, and make run eval easy to find. Map loading, empty, failure, permission and success states before you create editable high-fidelity screens.

Open Figr