GitHub user cerebrixos created a discussion: Airflow-native governed AI endpoint example?
I had opened a small DAG how-to PR for calling Tuning Engines from Airflow, but I want to make sure any follow-up is useful to Airflow users rather than a generic gateway listing. Would an Airflow-native example be a better fit if it demonstrates: - storing the inference key in an Airflow Connection or Secret Backend - using Airflow retries/timeouts/pools to control LLM call reliability and spend - passing `run_id` and `request_id` from an Airflow task into the governed inference request - returning only safe summaries/metadata through XCom - linking Airflow task logs to Tuning Engines traces, policy decisions, approvals, token usage, and cost The pattern would keep Airflow as the scheduler/orchestrator and Tuning Engines as an OpenAI-compatible governed inference endpoint. I can open a replacement PR if a cookbook/how-to like that fits the docs, or keep it external if the project prefers not to include third-party endpoint examples. GitHub link: https://github.com/apache/airflow/discussions/67780 ---- This is an automatically sent email for [email protected]. To unsubscribe, please send an email to: [email protected]
