Reading Invoice

Pulls structured data from invoice images, handling both PDF and image formats. The dual context verification approach is the interesting bit here: it reads the same invoice twice in parallel using separate subagents and flags discrepancies for you to resolve, which catches OCR mistakes that would otherwise slip through. Extracts vendor names, totals, tax amounts, line items, and Japanese qualified invoice numbers. Handles Japanese dates including era conversions (Reiwa to Gregorian). If you're processing invoices in a region where tax rates vary or invoice accuracy matters for compliance, the verification step saves manual checking time. Falls back to single read with manual confirmation if subagents aren't available.