Question

A GenAI engineer has deployed a RAG chatbot to a Databricks Model Serving endpoint. The compliance team now requires that every request and response be captured and stored in Unity Catalog so they can later audit prompts, retrieved context, and generated answers, and run offline quality analysis. The engineer wants the lowest-effort, Databricks-native approach that does not require adding custom logging code inside the chain. What should the engineer do?

Accepted Answer

Enable inference tables (payload logging) on the serving endpoint so requests and responses are automatically written to a Delta table in Unity Catalog.

Answer

Add Python logging statements inside the chain's predict function to append each prompt and response to a Delta table using a Spark write.

Answer

Configure the endpoint's autoscaling settings to capture traffic metrics and export them to a Unity Catalog volume.

Answer

Enable MLflow autologging in the model's logging code so every serving invocation is tracked as an MLflow run.