Project Overview
Situation
California PAGA filing and settlement research was a manual, error-prone process spanning email intake, PDF retrieval, data extraction, and reporting — unsuited to long-running, unattended processing.
Task
Design a reliable production pipeline covering Outlook intake, search/data collection, PDF retrieval, AI-assisted extraction, validation, Airtable persistence, CSV generation, and email delivery.
Action
Built an orchestrator/dispatcher/worker architecture in n8n with request IDs, record claiming, locks, concurrency controls, daily caps, retries, heartbeats, and PENDING/PROCESSING/COMPLETED/ERROR state management; implemented multi-agent LLM extraction (numeric settlements, litigation/mediation context, valuation/exposure); integrated Power Automate Desktop, Browserless, and ShareFile APIs; hardened against malformed PDFs, stale claims, and service-account issues.
Result
Production legal-research automation with full request tracing and operational error handling — evolved from a linear flow into a resilient, observable dispatcher/worker system.