AI Services • Posted 2 months ago
We receive about 300 documents a day by email — purchase orders, invoices, delivery notes, complaints — and a person sorts them into folders. We want automatic classification and routing, with a confidence threshold below which it goes to a human rather than guessing. Mixed quality: some are clean PDFs, some are photographs of paper taken at an angle. Handling the bad ones well is the actual job. Please tell us how you would measure this in production, not just on a test split.
