All posts
🧾
The true cost of manual data entry for small businesses
Three hours a week sounds manageable. Multiply it by 52 weeks, factor in errors and rework, and the number gets uncomfortable fast.
🤖
How we trained our document classifier on less than 2,000 samples
Getting useful ML classification without a massive labelled dataset requires some creative thinking. Here's what worked for us.
☁️
Bring your own storage: why we made Google Drive a first-class option
Vendor lock-in is a real concern for small businesses. We built Paperocket so your documents always live where you decide — not where we decide.
🔍
OCR in the real world: handling blurry receipts, bad lighting, and handwriting
Tesseract is great on clean PDFs. It's not so great on a photo taken from the back of a delivery truck at 5pm. Here's how we handle the hard cases.
📊
Getting ready for Xero sync: how to structure your document workflows
Before you connect your accounting platform, it's worth thinking about how your approval flow maps to your chart of accounts. A short guide.
🛠️
Why we chose FastAPI + Supabase over a more traditional stack
Speed of iteration, built-in RLS for multi-tenancy, and a storage layer we didn't have to build ourselves. Here's the full reasoning.

Get new posts in your inbox.

No spam. Just occasional posts on documents, automation, and building a company in New Zealand.