Explain the score
Keep criteria and thresholds visible so people can understand what a result means.
E-VAL · WORK IN PROGRESS
Make AI evaluation
easier to interpret.
A workspace for testing prompts and understanding evaluation results, with a comparison experience in development.
THE DESIGN QUESTION
A score is a starting point.
E-Val explores how practitioners can understand why an AI response passed or failed, then compare prompt or model configurations under consistent conditions.
My starting hypothesis is that useful evaluation tools need to connect summary scores to individual test cases and the criteria behind them. User research has not yet tested that framing.
MY FOCUS
From results to decisions.
I am reviewing the existing workflow and shaping the evaluation and comparison experience. My contribution is UX and product design; the existing implementation is the starting point for this work.
Make failures inspectable
Connect a summary to the response and supporting evidence behind it.
Compare under clear conditions
Make changes in prompts, models, and evaluation settings explicit.
These are proposed design directions. Their usability has not yet been evaluated.
STATUS AND NEXT STEPS
An early product, with research ahead.
The supplied project record documents an early interface with datasets, runs, rubrics, and stored DeepEval results. A/B comparison is in development. Other framework integrations and adversarial testing capabilities remain unverified.
Next, I will observe practitioners evaluating real outputs, prototype the path from scores to individual cases, and test whether users can interpret results and spot misleading comparisons.
Scores describe selected tests and criteria. They do not establish overall model quality or safety. Product screens and validated findings will be added as the work progresses.