Thrive Invoice Imports

At Shopventory, I worked on a feature that let customers create purchase orders by simply uploading a PDF or JPG of a paper invoice, rather than entering line items by hand. The system used Google Vertex AI to extract structured data from the document: vendor and order metadata, plus a full table of line items with quantities and prices. It parsed the documents accurately even from a scanned or photographed paper invoice rather than a clean digital file. I worked on the frontend experience end-to-end: the upload flow, the review screen where users could correct anything the model got wrong before committing, and the UX around confidence and error states when extraction was incomplete or ambiguous.

Once the model produced structured line items, our backend matched each one against existing product data. A purchase order could then be generated with products already linked. This saved customers time by replacing their data entry work with a new review-and-confirm workflow.