Loading the studio…
Northgate Accounting · AI Automation / Software
Built an OCR + LLM pipeline that ingests messy client invoices, classifies them, extracts line items, and pushes structured data straight into QuickBooks Online.
62 hr/mo
Manual data-entry hours saved
9 min → 35 sec
Invoice processing time
−48%
Error rate vs human entry
Bookkeeping team was drowning. Clients emailed PDFs in 14 different formats — some scanned, some screenshot, some retyped — and a junior bookkeeper retyped each one into QuickBooks. End of every month was a fire drill.
Built a classifier that handles 14 vendor templates plus a generic fallback.
Layered an LLM extraction step for line items, with confidence scores per field.
Added a human-in-the-loop review queue only for low-confidence rows.
Pushed validated invoices straight into QBO via the API with audit trail.
62 hours of monthly data entry gone. The junior bookkeeper is now doing client advisory work that bills at 4× the hourly rate of typing.
“Our month-end used to be a survival exercise. Now it's a Wednesday.”