Pipeline Copilot

Pipeline Copilot. When a data pipeline breaks at 2 AM, the on-call engineer gets a wall of logs, a failed SQL query, and very little else. Finding the root cause can take hours, and the business dashboards stay wrong until someone does. Pipeline Copilot takes a failed pipeline run, the error log, the SQL, the table schemas, and recent changes, and returns three things in under a minute: the most likely root cause in plain English, the evidence from the logs that supports it, and a draft fix (corrected SQL or config) the engineer can review and apply. It covers the failures data teams hit most: schema drift, bad upstream data, permission changes, and silent data quality explosions. The diagnosis engine is deterministic and runs locally with no API key; classification and evidence always come from the local engine, so it never invents a root cause. The Meta Model API layer (Muse Spark, verified live) rewrites the summary and reviews the draft fix when a key is set. Built solo in Python (standard library only) with six realistic sample failed runs from Snowflake and dbt-style pipelines. Run it with: python3 server.py, then open http://localhost:8000.