I need a small GPT-powered utility that can open a batch of manufacturing blueprint PDFs—portrait or landscape—extract four fields (Part Name, Model No., Drawing No., Revision No.) and push the results straight into an Excel file. Layouts are not guaranteed to be identical and the data might appear anywhere on the sheet, so the model has to be smart enough to locate the fields even when they drift outside a tidy title block. Core workflow I am envisioning • User clicks once, selects any number of PDFs. • Tool runs OCR if required, reads every page, finds the four data points, writes them to a single spreadsheet row per drawing (columns: Part Name, Model No., Drawing No., Revision No.). • Each source file is then renamed to “<DrawingNo>_<RevisionNo>.pdf”. Because these are manufacturing blueprints, accuracy is critical: if a field is missing or doubtful the script should flag it rather than guessing. Python with pdfplumber / PyMuPDF, pytesseract, and the OpenAI API feels right, but I am open to whatever stack gets reliable results and keeps runtime reasonable on a standard Windows workstation. Acceptance will be based on: 1. A compiled or easily runnable script plus source code. 2. Correct extraction and renaming on a mixed test set of 50 PDFs containing both portrait and landscape pages, with at least 98 % field accuracy. 3. One-click batch processing UI (simple desktop window or command line with drag-and-drop is fine). 4. Delivered Excel file matching the sample format. If you have prior experience parsing variable-layout technical drawings, please let me know—otherwise just outline how you plan to tackle free-floating text blocks and possible OCR noise, and we can get started.