Skip to content

Evaluation: Update prompt iteration flow #1072

Description

@AkhileshNegi

Is your feature request related to a problem?
Prompt iteration currently works on the old evaluation flow: LLM-as-judge correctness and cosine similarity results were used to generate prompt improvements. V2 evals now produce three metrics — adherence to ground truth, adherence to prompt, and adherence to knowledge base — each with a score and judge reasoning. The existing endpoint neither consumes the new results shape nor reasons over the per-metric scores.

Describe the solution you'd like

  • Add a v2 prompt iteration endpoint that consumes the three-metric eval results instead of correctness + cosine similarity
  • Use both the score and the judge reasoning of each of the three metrics when generating prompt improvement suggestions
  • Update the recommendation prompt accordingly
  • Keep the API extensible for future recommendation types (knowledge base gaps, model changes) — prompt recommendations first

Metadata

Metadata

Assignees

Labels

enhancementNew feature or request

Type

No type

Projects

Status
In Progress

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions