Eval (LLM-as-judge)
An LLM scores LLM responses against 5 built-in criteria (plus your own) and surfaces drift versus past runs.
Demo: Run #42 scores (= 12 responses × 5 criteria)
| Helpfulness | Accuracy | Relevance | Safety | Conciseness | |
|---|---|---|---|---|---|
| call_8f12 | 5 | 5 | 4 | 5 | 5 |
| call_4a01 | 4 | 4 | 5 | 5 | 4 |
| call_b3c5 | 3 | 4 | 4 | 5 | 3 |
| call_d770 | 5 | 4 | 5 | 5 | 5 |
| call_e2a8 | 4 | 4 | 4 | 5 | 4 |
Mean score4.40vs last run +0.18
Trend: improving5 default criteria
Helpfulness / accuracy / relevance / safety / conciseness. Ready to use.
Custom criteria
Add "matches our brand voice", etc. (Pro+). Each criterion's scoring rubric is editable.
Compare runs
See mean score and failed-count delta vs the previous run in one view.