Automation workflow · 2025 · n8n automation, AI & RAG, Integrations
OCR document extraction pipeline
A document inbox that converts incoming images into structured records and moves each source through an idempotent processed state.
Problem
Manual transcription is slow, while pure OCR returns text rather than the consistent fields an operation needs.
My role
Workflow design, OCR, schema extraction, Drive/Sheets integration, and idempotency
Stack
How the work moves
- Step 01Drive inboxNew image documents
- Step 02OCRRaw text extraction
- Step 03SchemaLLM field normalization
- Step 04RecordSheets row + processed file
The business problem
OCR returns text, but an operation usually needs consistent fields. Manual transcription also needs a clear inbox so the same document is not processed twice.
Workflow design
A Google Drive folder acts as the inbox. Tesseract extracts raw text, then an LLM information extractor maps it to a defined schema. The workflow appends the structured record to Google Sheets and moves the source into a processed folder.
The folder transition is the idempotency boundary: new work and completed work remain visible in familiar tools.