Python automation developer

About us We are a licensed apparel sourcing and consulting company based in Spain. We produce branded, licensed garments for major European and Latin American retailers under licences from entertainment and character brands. We want to automate an internal workflow that currently runs manually. The goal When a purchase order arrives, a member of our team manually reads it, fills in a label-instruction file, applies the relevant brand-licence and retailer-specific rules from memory, and produces an Excel instruction file for the factory. We want a system that automates the data extraction and rule application end-to-end and outputs a finished, compliant instruction file — while leaving our licence approvals and licensor submissions as manual human steps. What the system must do 1. Read incoming purchase orders. Ingest POs received by email in several formats — Excel (one retailer) and PDF (several retailers, with different layouts) — and extract the label data automatically (EAN codes, sizes, composition, price, etc.; roughly 150–200 data points per order). Missing or invalid fields must be flagged for human review. No silent errors. 2. Validate the data. EAN13 codes must be check-digit validated before anything is written to the output. Invalid codes are blocked and flagged. 3. Apply brand-licence rules automatically. Each label must carry the correct copyright/credit line for its brand, with the correct year. Any licence without a confirmed credit line must be held and flagged — never guessed. 4. Apply country and retailer rules automatically. Language requirements (some brands restrict to specific languages), OEKO-TEX where required, and retailer-specific specifications (hanger colour, alarm/security codes, logo rules) must be applied by the system rather than from a person's memory. 5. Generate the finished label file. Output is an .xlsx instruction file in our existing template format. Critical: EAN13 and DUN14 codes must be preserved as text (not numbers), images must be embedded, and every label must already be compliant with the correct licence and country/retailer rules. 6. Editable rule library (key requirement). Our non-technical staff must be able to add a new licence, change a credit line, or update a country/retailer rule directly — without needing a developer each time. Please propose how you would implement this (e.g. a no-code data store such as Airtable or Google Sheets that the system reads, or a simple admin interface). In scope / out of scope In scope: everything from reading the order through generating the finished instruction file (items 1–6 above). Out of scope: licensor portal submissions and human licence approvals — these stay as manual steps and are not part of this build. Technical notes (open to your recommendations) We expect a Python-based solution. Because the incoming PDF layouts vary and change over time, we favour an LLM-based extraction approach (e.g. Claude or OpenAI) over a separate hardcoded parser per format. If you would approach it differently, tell us why. Excel output handled with openpyxl or equivalent, with strict text-formatting of codes and embedded images. Email ingestion via IMAP or Microsoft Graph (we use Outlook). What we will provide Sample purchase orders in each format, our product sheets, the current Excel template, the brand-licence credit-line list, and the country/retailer rule set. Proposed milestones Extraction prototype (paid). Reliable extraction demonstrated on two or three of our real purchase-order files. This is the make-or-break step and we want to validate it before committing to the full build. Data validation, licence rules, and country/retailer rules. Excel output in our template format. Editable rule library. Integration, testing on real orders, documentation, and handover.