AI Invoice-to-PO Matching

Problem / Context

Manufacturers manually matching supplier invoice PDFs/images against purchase orders — slow and error-prone AP work.

What it does

• Extracts structured data from invoice PDFs/images via OpenRouter LLM
• Matches against POs, flags discrepancies, routes exceptions for human approval
• Dashboard, email digest, per-invoice AI cost tracking
• Demo accounts (admin/approver/viewer) and seeded PO for demo flow

Engineering notes

Laravel 12 API + React/Vite (TS) frontend + n8n orchestration + Mailpit; PostgreSQL 16; Docker Compose. Default model openai/gpt-6-luna-pro (~$0.10/M input tokens), configurable via AI_MODEL.

Outcome / Value

AI-powered AP automation showcase for LLM extraction + human-in-the-loop workflows.