VeriClaim

Expense review makes finance teams read receipts, interpret policies and chase approvals across disconnected tools. VeriClaim brings these steps into one inspectable workflow, helping teams automate routine checks while keeping exceptions visible. Built with Flask, VeriClaim drives Evolus through its REST API. A document extractor reads receipt fields, and a policy reviewer assesses business-purpose and category concerns. The reviewer is connected to a policy knowledge library and receives a numbered policy reference. Inference uses the team's AMD-hosted model connection. Python independently applies the fictional USD policy's spending limits, receipt-age rules and approval thresholds. Eligible claims can be auto-approved, unresolved cases go to a human reviewer, and unreliable extraction is marked Needs attention. The app preserves receipts, extracted fields, review findings, rule checks and available Evolus execution IDs. Unsupported policy quotations are removed and flagged for review. Our key distinction is the separation of AI interpretation from deterministic decisions. Finance staff can inspect the evidence and record an approval or rejection instead of relying on an unexplained model answer. The intended business value is less repetitive checking and clearer handoffs; savings have not yet been measured. The prototype was tested with 60 fictional receipts through real Evolus calls and 74 offline regression tests. Docker configuration and setup instructions are included. OCR and optional fields can still be misread, so human review remains essential. VeriClaim demonstrates expense-review automation; it does not make payments or claim production accuracy or fraud detection.