
InvoiceLeak helps finance and procurement teams detect silent margin leaks before they become recurring losses. Users upload invoice PDFs, contract PDFs, and usage CSV files; the system extracts structured information, runs deterministic checks, investigates inconsistencies across documents, verifies the findings, and generates dispute-ready actions for vendors. The model design is intentional. Gemini 2.5 Flash is used for the extraction and verification stages because Google positions it as its best price-performance model for low-latency, high-volume reasoning tasks, making it effective for quickly processing many documents and checking outputs efficiently. Gemini 2.5 Pro is used for the investigator stage because Google describes it as its most advanced model for complex reasoning and analysis across large datasets, making it better suited for comparing contract clauses, invoice details, and usage records before producing evidence-backed findings. In other words, InvoiceLeak uses Flash where speed and efficiency matter most, and Pro where deeper reasoning creates the most value.
19 May 2026