Skip to content

Browser Automation Agent for Back-Office Ops

A vision-and-DOM agent that automates legacy portal workflows with structured retries and human-in-the-loop fallback.

OpenAI APIPlaywrightFastAPITool Calling
Abstract charcoal and magenta cover illustration for the Browser Automation Agent project, showing layered portal frames wired by fuchsia action pathways.

Highlights

  • Hybrid vision + accessibility-tree targeting resilient to layout drift.
  • Deterministic step recorder with LLM repair on failure.
  • Full trace video and structured audit log per run.

Outcomes

  • Automated 12 back-office workflows previously done by hand.
  • Cut per-workflow labor cost by 82% at steady state.

Stack

  • OpenAI API
  • Playwright
  • FastAPI
  • PostgreSQL
  • Redis