Description
We are building an automation platform for a services firm, on top of the systems they already run. This contract covers the first workstream: an end-to-end pipeline that captures supplier invoices, extracts the accounting data, and posts entries into the client's ERP for human approval. Architecture, data model and acceptance criteria are written by our technical lead before each milestone starts, so you always work against a written spec and a written definition of done. What the pipeline does - Ingests invoices from three sources: a shared Microsoft 365 mailbox (read-only), a hot folder on a NAS, and manual upload from a web dashboard. - Reads multi-format documents — native PDF, scanned PDF, images — including files that contain several invoices in one document. - Extracts issue date, supplier, amounts (gross / net / tax) and tax codes; identifies the end client through a documented key hierarchy (tax number → registered contact → name match → manual queue). - Archives the original file on the NAS, in the folder structure and naming conventions already in use. - Creates the accounting entry in the ERP through its API, flagged as pending review. Nothing reaches the ledger without human validation. - Sends anything unreadable, unidentified or low-confidence to a review queue, together with the original file, the extracted fields and the reason it was held. - Logs every action with author, timestamp and source. User corrections are stored and fed back into extraction quality