AI Workflow Automation

Maintainable n8n Pipelines and a Read-Only SQL Agent

Repeat automation work moved scripted data flows into n8n and added a database-question agent with an explicit SQL execution boundary.

Original client-supplied PostgreSQL table overview used as project input; not an agent-result screenshot
Original client-supplied PostgreSQL table overview used as project input; not an agent-result screenshot

The problem

The team needed a multi-step data pipeline maintained as inspectable workflows and a way for operators to ask questions of Postgres. Moving scripts into visual nodes had to preserve insert, update and delete handling. The question-answering agent needed a separate read-only execution path.

What I built

  • n8n workflows for collection, comparison, enrichment and record synchronization.

  • Separate insert, update and delete routes with shared error handling.

  • AI enrichment and vector-index synchronization with confidence-based review.

  • A chat agent that loads the database schema and calls an SQL tool.

  • SELECT-only validation and result limits in the executor path.

  • Postgres conversation memory and handoff documentation.

These are related engagements in one case study. The SQL restrictions apply to the question-answering agent, not to the separate data-synchronization workflow.

The outcome

The result is an inspectable n8n pipeline and a constrained way to ask operational database questions. Uncertain enrichment and empty query results remain visible. The local archive includes evaluation plans, not a scored agent-accuracy result. No standalone Python service or measured savings figure is claimed from these artifacts.

Working on something similar?

Discuss a similar project →