AI Agents That Parse Invoices and Purchase Orders from Gmail
AI agents fetch PDFs and images straight from Gmail and Drive, extract every field with OCR + LLM parsing, and write clean structured data to Sheets - without anyone touching an inbox.
Someone opened every invoice and PO email by hand, downloaded the attachment, and retyped the numbers into a tracking sheet - line items, amounts, vendor details, all re-entered manually with no consistency check.
An AI agent watches the inbox, pulls PDFs and images straight from Gmail and Drive, runs them through OCR (DeepSeek OCR, Azure Document Intelligence, or LlamaParse) and OpenAI extraction, and writes the structured result straight into Google Sheets - with a filter step so the same email never gets processed twice.
Stack
How It Works
Watch the inbox
The AI agent monitors Gmail and Drive for new invoice or PO attachments - PDFs and images alike.
Extract with OCR + LLM
The document is parsed with a specialized OCR service, then OpenAI extracts and structures every field: vendor, line items, totals, PO number.
Cross-check against tools
The agent can query a vector database (Supabase or Pinecone) or external APIs to validate vendor details or match against known records.
Write to Sheets and file the PDF
Structured data lands in Google Sheets; the original PDF is filed into a systemized Drive folder.
Dedup guard
A filter step checks whether that exact email has already been processed, so re-runs never create duplicate rows.
Results
saved on manual data entry
duplicate records
attachments auto-filed
The Workflow
