Databricks Mlflow Evaluation

MLflow 3 GenAI Evaluation

Scope vs upstream mlflow/skills

The OSS mlflow/skills repo ships agent-evaluation and related skills (instrumenting-with-mlflow-tracing, analyze-mlflow-trace, retrieving-mlflow-traces, querying-mlflow-metrics) that cover the generic MLflow GenAI evaluation workflow — mlflow.genai.evaluate(), scorers/judges, datasets, tracing setup, and the 5-step evaluation loop.

This skill layers Databricks-specific patterns on top of that workflow rather than restating it. Use this skill when you need any of:

• Unity Catalog trace ingestion — production traces written into UC tables, log-based monitoring (patterns-trace-ingestion.md).
• MemAlign judge alignment via UC SME labeling sessions — aligning custom judges against domain-expert feedback collected in Databricks (patterns-judge-alignment.md).
• optimize_prompts() GEPA loop — Databricks' automated prompt-optimization driver running on a UC dataset (patterns-prompt-optimization.md).
• Databricks-flavored scorer/dataset patterns — UC-table-backed datasets, tagging traces in the Databricks UI for inclusion (patterns-datasets.md, patterns-scorers.md).

For everything else — generic mlflow.genai.evaluate() calls, scorer authoring patterns, dataset creation outside Databricks, MLflow tracing setup that isn't UC-table-bound — the upstream mlflow/skills/agent-evaluation skill is the canonical source and is kept current by the MLflow team.

Before Writing Any Code

1. Read GOTCHAS.md - 15+ common mistakes that cause failures
2. Read CRITICAL-interfaces.md - Exact API signatures and data schemas

End-to-End Workflows

Follow these workflows based on your goal. Each step indicates which reference files to read.

Workflow 1: First-Time Evaluation Setup

For users new to MLflow GenAI evaluation or setting up evaluation for a new agent.

Workflow 2: Production Trace -> Evaluation Dataset

For building evaluation datasets from production traces.

Workflow 3: Performance Optimization

For debugging slow or expensive agent execution.

Workflow 4: Regression Detection

For comparing agent versions and finding regressions.

Workflow 5: Custom Scorer Development

For creating project-specific evaluation metrics.

Workflow 6: Unity Catalog Trace Ingestion & Production Monitoring

For storing traces in Unity Catalog, instrumenting applications, and enabling continuous production monitoring.

Workflow 7: Judge Alignment with MemAlign

For aligning an LLM judge to match domain expert preferences. A well-aligned judge improves every downstream use: evaluation accuracy, production monitoring signal, and prompt optimization quality. This workflow is valuable on its own, independent of prompt optimization.

Workflow 8: Automated Prompt Optimization with GEPA

For automatically improving a registered system prompt using optimize_prompts(). Works with any scorer, but paired with an aligned judge (Workflow 7) gives the most domain-accurate signal. For the full end-to-end loop combining alignment and optimization, see user-journeys.md Journey 10.

Reference Files Quick Lookup

Critical API Facts

• Use: mlflow.genai.evaluate() (NOT mlflow.evaluate())
• Data format: {"inputs": {"query": "..."}} (nested structure required)
• predict_fn: Receives **unpacked kwargs (not a dict)
• MemAlign: Scorer-agnostic (works with any feedback_value_type -- float, bool, categorical); token-heavy on the embedding model so set embedding_model explicitly
• Label schema name matching: The label schema name in the labeling session MUST match the judge name used in evaluate() for align() to pair scores
• Aligned judge scores: May be lower than unaligned judge scores -- this is expected and means the judge is now more accurate, not that the agent regressed
• GEPA optimization dataset: Must have both inputs AND expectations per record (different from eval dataset)
• Episodic memory: Lazily loaded -- get_scorer() results won't show episodic memory on print until the judge is first used
• optimize_prompts: Requires MLflow >= 3.5.0
• UC model service judge models: databricks:/system.ai.<model> judge/scorer URIs need a version floor and can hit a 404 (old client) or 403 (legacy endpoint name) -- see GOTCHAS.md -> "Custom Judge Model Format" for the version, exact error text, and fix

See GOTCHAS.md for complete list.

Related Skills

• databricks-docs - General Databricks documentation reference
• databricks-model-serving - Deploying models and agents to serving endpoints
• databricks-agent-bricks - Building agents that can be evaluated with this skill
• databricks-python-sdk - SDK patterns used alongside MLflow APIs
• databricks-unity-catalog - Unity Catalog tables for managed evaluation datasets