Phoenix enables teams to identify data points for fine-tuning based on production performance and user feedback, such as high hallucination rates, and supports exporting these datasets for retraining. However, Phoenix does not natively provide built-in automation to push alerts or trigger retraining workflows directly to CI/CD or orchestration tools (e.g., via webhook, Airflow, or SageMaker integration); such automation must be implemented externally by monitoring Phoenix metrics and programmatically exporting data as needed according to <https://arize.com/docs/phoenix/user-guide#fine-tuning|Phoenix documentation>, <https://arize.com/docs/phoenix/~gitbook/pdf#B%3Aaf|be8694ec>, and <https://arize.com/docs/phoenix/~/changes/v6Zhm276x8LlKmwqElIA/concepts/llm-observability#testing-staging|c26af0d7>.
Would you like more detail on how to set up external automation for closed-loop retraining with Phoenix?
Sources:
- <https://arize.com/docs/phoenix/~/changes/v6Zhm276x8LlKmwqElIA/concepts/llm-observability#testing-staging|Phoenix LLM Observability: Testing/Staging & Production>
- <https://arize.com/docs/phoenix/user-guide#fine-tuning|Phoenix documentation>
- <https://arize.com/docs/phoenix/~gitbook/pdf#B%3Aaf|be8694ec>
