Descriptions:
Stephanie Nyarko builds a receipt processing automation in n8n that turns a phone photo of a receipt into a structured bookkeeping record. The workflow accepts image uploads through an n8n form, normalizes the binary file with a Code node to prevent downstream naming errors, sends the image to OpenAI Vision for OCR and field extraction, then passes the result through an AI Agent node configured with a strict JSON schema to produce clean structured data containing merchant name, purchase date, total, tax, currency, category, and a needs-review flag.
A second Code node generates two derived values: a composite deduplication key combining date, merchant, currency, and total (so re-uploading the same receipt does not create duplicate records), and a clean human-readable filename. A Merge node synchronizes the parallel data streams before a Google Drive node uploads the original image under the new filename, and an n8n data table node performs an upsert using the dedup key to log the structured fields.
The tutorial is technically detailed and covers real engineering considerations: file size limits on n8n form triggers, why binary key normalization matters, how to configure structured output parsing inside an AI Agent node, and how the Merge node resolves multi-stream timing issues. The result is a searchable, categorized receipt archive ready for accounting workflows.
📺 Source: Stephanie Nyarko · Published January 13, 2026
🏷️ Format: Hands On Build
