How We Processed 600 Real Tax Invoices in 4.5 Minutes Using a Dual-Engine Gemini AI & Vertex AI Architecture

DiGitax built a dual-engine document processing pipeline that routes tax invoices through either Google Gemini AI for free daily ingestion or Google Cloud Vertex AI for high-volume bulk batches, cutting the processing of 600 Dominican Republic tax receipts from roughly 15 hours of manual entry to 4 minutes and 31 seconds. The company reports 100% field extraction accuracy on RNC, NCF and ITBIS fields, with Vertex AI throughput of 1,500 to 4,000 documents per hour at $0.0042–$0.0060 per document and Gemini ingestion at zero cost. Extracted documents pass through a four-filter DGII validation and depuration pipeline that flags duplicates and fiscal mismatches before reaching the general ledger.

Manual data entry remains the single biggest bottleneck for Certified Public Accountants CPAs and SMEs in Latin America and the Caribbean. In the Dominican Republic, preparing the monthly DGII Form 606 Purchases , Form 607 Sales , and IT-1 VAT/ITBIS declarations for a mid-sized business with 600 monthly receipts typically consumes 15 hours of manual typing and tax ID validation . At DiGitax Suite https://digitax.do Corporate Profile — DiGitax LLC https://digitax.do/company , we engineered a dual-engine AI document processing pipeline that cuts those 15 hours down to 4 minutes and 31 seconds with 100% field extraction accuracy —while keeping daily AI ingestion free for accountants. Here is the architecture and the full benchmark data from our 600-Invoice AI Case Study Gemini AI vs. Vertex AI https://digitax.do/escaneos-ia-gemini-vertex-digitax . Most accounting scanners fail in production for two reasons: B02 , and duplicate PDFs. If an OCR tool blindly dumps all of that into the ledger, the accountant spends more time cleaning up the mess than typing it manually. 0 Credits vs. 6 Credits/doc Instead of forcing every document through an expensive enterprise endpoint, DiGitax https://digitax.do routes documents across two specialized engines backed by a Unified Non-Expiring Prepaid Credit Wallet https://digitax.do/creditos-lotes-precio-digitax : | Metric / Capability | Track 1: Google Gemini AI 0 Credits / $0.00 Free | Track 2: Google Cloud Vertex AI 6 Credits / doc | |---|---|---| | Target Workload | Continuous daily ingestion WhatsApp, mobile app, email | End-of-month bulk closures & multi-page PDF batches | | Throughput Speed | 45 documents / hour +20 free scans/day per Google key | 1,500 to 4,000 documents / hour ~4.5 min per 300 docs | | Multi-Invoice PDF Splitting | Up to 3 invoices per PDF | Unlimited splits 50–100 page stapled PDFs automatically | | Extraction Accuracy RNC, NCF, ITBIS | 100% | 100% | | Cost per Document | $0.00 USD 0 Credits | $0.0042 – $0.0060 USD 6 Credits | Declarados vs. Descartados Before any extracted document touches the General Ledger, it passes through our 4-Filter DGII Validation & Depuration Pipeline https://digitax.do/ingesta-inteligente-depuracion-facturas-606 flujo-bandeja-fiscal : Fiscal & Accounting Track Left Rail : Procesados → Contabilizados → RNC Activo and voucher sequence NCF Serie B / e-CF Serie E under Control & Depuration Track Right Rail : Pendientes → Incidencias → RNC + NCF duplicates are automatically diverted to