OrderGuard AI: Invoice Resolution Assistant

Procurement and accounts payable teams spend time reconciling purchase orders, supplier invoices, and goods-received notes. OrderGuard AI turns that manual three-way check into a clear, evidence-based resolution workflow. Users upload three PDFs. A Python verification engine extracts the document fields and compares quantities, unit prices, references, suppliers, and totals. It highlights mismatches and reports preliminary financial differences without treating them as confirmed losses. An Evolus resolution agent examines the structured findings, explains uncertainties, recommends responsible next actions, and supports supplier clarification drafting. The public demo walks through a synthetic case with 50 units ordered, 50 received, and 60 invoiced. OrderGuard identifies the 10-unit discrepancy and a potential USD 200 difference. A reviewer can prepare an editable supplier message, record a decision, upload a revised invoice, rerun verification, and confirm resolution only after the new document passes the configured checks. Built with Python, Streamlit, PyMuPDF, an Evolus AI agent, and an AMD Radeon API analysis option. We also built and tested a six-block Evolus automation with an asynchronous API handoff; its public trigger is restricted to protect credits. OrderGuard never automatically sends supplier emails, approves payments, or closes cases without human confirmation. This is a synthetic-data hackathon MVP, not a production accounting system.