Thinking on data engineering, AI infrastructure, and what actually works in production.
A team built twelve data pipelines in Airflow using Claude Code. The dashboards were green, leadership was happy, and nobody asked what was inside. Then the engineer who built it left.
Read →A marketing analytics pipeline was built around an AI that read PDFs, Excel files, and screenshots to extract campaign data. The model worked fine. The problem was that 'fine' meant something different every run.
Read →