Field notes, not tutorials
Airflow, dbt, BigQuery, GCP — and AI where it belongs: inside the pipeline. Everything here comes from projects I've actually built and shipped.
World Bank API → Airflow → DuckDB → dbt. What building on live data (BOM quirks, missing countries and all) teaches you that synthetic datasets never will.
Read articleHow persisting dbt artifacts to GCS turns a full-pipeline retry into a surgical re-run of failed models on Kubernetes.
Read articleWhen every new pipeline is a copy-pasted Python file, your Airflow repo rots. Config-as-code with validation at import time fixes it. This is how I built one.
Read articleData engineering and ML meet in production. How a nightly credit-scoring DAG uses the Population Stability Index as a quality gate, so a broken feature never reaches a lending decision.
Read article