Skip to content

Support batch insights in BatchEvaluationRunner #678

Description

@jariy17

BatchEvaluationRunner does not expose the batch insights workflow.

Requested behavior

Support requesting batch insights through BatchEvaluationRunner, with job polling and access to the resulting insights. The original report identifies these StartBatchEvaluation insights:

  • Builtin.Insight.FailureAnalysis
  • Builtin.Insight.UserIntent
  • Builtin.Insight.ExecutionSummary

The workflow should enforce the API constraint that insights and evaluators cannot be requested in the same job.

Why this matters

The reporter uses insights to cluster failures and identify root causes that can be addressed through prompt changes. Without SDK support, the evaluation improvement workflow requires custom orchestration outside BatchEvaluationRunner.

Open design question

Confirm whether the runner's dataset-based workflow is compatible with insights and which inputs are supported. Use that finding to determine how to expose insights through the runner, or whether a separate insights runner is needed. This compatibility question remains unresolved.

Original report environment: bedrock-agentcore 1.23.0, Python 3.13, us-west-2.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or request

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions