Ope Buhali

The catalog kept filling up with duplicate products. Same item, different photo, new SKU. It made everything harder to manage.

I used CLIP to embed product images and FAISS to search for similar ones across the catalog. Added OCR for spec sheets so text could also be matched. When a new product comes in, the system flags potential duplicates for admin review before creating a SKU.

Duplicates get caught early instead of cluttering the catalog later. The system stays out of the way until it finds something worth checking.

Finance staff had to read transfer receipts from images, type everything into Google Sheets, then match it all against bank mutations at month end. Doing this for hundreds of receipts every month was exhausting.

Split it into two n8n workflows. The first takes uploads, hashes each file, stores it in Google Drive, and queues it in Google Sheets. The second runs on a schedule and reads each receipt with Tesseract first, then OpenSpace OCR with per-bank parsers. An AI agent is the last resort for blurry receipts, and its results are flagged for manual review.

A task that used to take hours now finishes in minutes. No more daily overtime just to keep up with receipts. Some blurry images still need manual review, but the repetitive typing is gone.

We get thousands of shipping receipts daily. Each courier uses a different format. Reading and matching them all by hand is exhausting.

Built with n8n on a 'you build, you use' approach: one template per courier, added only when needed. Text comes from the PDF, with Tesseract as fallback for scans. A template scorer picks the courier by keywords and receipt-number patterns, and the highest score decides the route. Each route has its own isolated parser that handles slash, dash and pipe item formats, with AI as fallback for anything unknown.

Over 35,000 receipts processed so far with zero read errors. Users don't pick a courier. Everything goes through one input and the system figures out the route.