.png&w=256&q=75)
EMAN ABBAS@Eman_Abass
🤝 Top Collaborators
.png&w=640&q=75)
🤓 Latest Submissions
.png&w=828&q=75)
Invoice Anomaly Desk
Finance teams still check supplier invoices by hand. That is slow and mistakes are costly: duplicate invoices get paid twice, amounts go above the purchase order, and fake "we changed our bank account" messages lead to fraud. Invoice Anomaly Desk is a web app with an AI agent behind it. A user uploads invoices as text PDF or CSV. The agent extracts the fields with a model running on Vultr Serverless Inference. Then plain code runs seven checks: duplicate invoice, changed bank details, amount above the purchase order, missing purchase order, price outlier against the vendor's history, math or tax error, and unknown vendor. Each invoice gets a risk score and is routed to auto-clear, review, or block. For every flag, the model writes a short plain-English explanation using only the evidence found by the code, and drafts a polite follow-up email to the supplier. A human reviewer approves, rejects, or sends the query. Nothing is paid or sent without a person, and every step is saved in an audit log. The backend is built with Node.js, Express and MongoDB, and is packaged with Docker for deployment on a Vultr Cloud Compute VM. A synthetic dataset with planted anomalies lets us measure how many problems the agent catches. Our goal is a lightweight, low-cost tool for small companies and bookkeepers
8 Nov 2026
