What problem does it solve?
This Skill eliminates the manual, error-prone work of tracking, running, and comparing AI model experiment results, making it easy to benchmark model performance and identify regressions without scattered spreadsheets or ad-hoc scripts.
Core Features & Use Cases
- Experiment CRUD Operations: Create, list, retrieve, and delete Arize experiments tied to specific versioned datasets.
- Run Analysis & Comparison: Export experiment runs, compare evaluation scores across model variants, and identify performance regressions between test versions.
- Use Case: For example, you can run a baseline experiment for your current production model, test a new prompt variant, and quickly compare correctness and relevance scores across both runs to validate improvements.
Quick Start
Use the arize-experiment skill to create a new experiment named 'claude-v3-test' against your latest customer support dataset, run inference on all examples, and compare its correctness score to your existing gpt-4o baseline experiment.