databricks-mlflow-evaluation

To evaluate GenAI agents using MLflow, you run mlflow.genai.evaluate() on a prepared evaluation dataset to score agent responses against configured metrics like Correctness and Safety.

Trace ingestion in MLflow GenAI evaluation workflows involves integrating production traces into experiments to build ground-truth datasets and enable ongoing agent monitoring.

You configure custom scorers for MLflow GenAI evaluations by defining and registering metrics such as Guidelines, RetrievalGroundedness, Correctness, and Safety to assess agent outputs.

Yes, you can build evaluation datasets from production traces in MLflow by structuring inputs as {'inputs': {'query': '...'}} and aligning them with reference materials as ground-truth expectations.

MLflow supports alignment workflows for GenAI agent evaluation by orchestrating scorers, evaluation datasets, and trace workflows to compare agent responses across multiple configured metrics.