Build Custom AI Agent to Process Emailed PDF Purchase Orders into ERP-Ready CSVs (Microsoft/Azure)

About us We’re a US medical supply manufacturer and distributor. We receive a steady daily volume of purchase orders by email as PDF attachments, from many different customers, each with its own PO format. Our customer service team currently keys these orders into our ERP by hand. We want to build our own AI agent to do this work, hosted in our own environment and owned by us. What we want built An AI agent that: 1. Monitors an order inbox (Microsoft 365 / Outlook / Front) and identifies emails that contain purchase orders. 2. Reads the PDF attachments and extracts the order data: PO number, customer, bill-to and ship-to, line items, quantities, units of measure, prices, requested dates, and shipping instructions (including third-party carrier account numbers when listed). 3. Handles real-world documents. Most are clean system-generated PDFs, but some are faxes, rotated or upside-down scans, multi-page documents with cover sheets and terms pages, and handwritten order forms. It must handle customer formats it has never seen before, so we are not looking for a per-customer template approach. 4. Matches order data against our ERP reference data (customers, ship-to addresses, item numbers including customer and legacy part numbers, units of measure, and prices), which we will provide. 5. Flags anything it isn’t confident about, such as unclear fields, price differences, unknown items or addresses, and missing PO numbers, so our reps know exactly what to check. 6. Outputs one CSV file per order in a defined format, to a location where our existing ERP import process picks it up. What’s already handled (not in your scope) - ERP import: our existing ERP has a CSV import process managed by our integration partner. They handle importing, duplicate order checks, and downstream review workflows. You will coordinate with them on the CSV specification. - Human review: every order is reviewed by a customer service rep before release, so the agent prepares orders and does not release them. Environment We’re a Microsoft shop (Microsoft 365, Outlook, Teams). We’d prefer the solution be built with Power Automate or similar Microsoft tools, running in Azure or on our own network. Tell us which tools and which AI model you’d use, and why. What we are NOT looking for - No third-party SaaS platforms or licensed document processing software. We want a solution we own and control, not a subscription-based product. - No third-party per-order, per-page, or per-transaction processing fees. We're comfortable paying for our own hosting, infrastructure, and AI model usage (such as Azure services). - No extended discovery or study phase. We have sample orders and a defined output format ready. We want a working prototype quickly. Ownership We will own everything: all code, documentation, and accounts, set up in our company’s name from day one. The solution must be documented well enough that another developer could maintain it. Timeline 1. Working prototype that processes our sample orders and produces correct CSV output, within about 2 weeks of starting. 2. Production-ready system running in our environment and tested with our reps, ideally within 6–8 weeks of starting. Next steps Applicants with strong proposals will receive a detailed project brief covering our systems, order volumes, sample orders, and the CSV specification, after signing an NDA. We’ll then meet by video call. Please propose your own price. Break it down by the two milestones above, and separately quote any optional ongoing support at an hourly or monthly rate.