AI Extractor for Technical Drawings
Budget: £1,500 – £3,000 GBP
I have a series of technical engineering PDFs that show individual parts—mostly simple profiles cut from a plate. From each drawing I need an AI model that pinpoints four specific attributes: Width, Length, Thickness and Material Grade. Because these values feed directly into our nesting and quoting software, they must come through with exact dimensional accuracy, not approximations.
You are free to choose your preferred stack—Python with OpenCV, Tesseract, TensorFlow or any vision + NLP combination—so long as the final solution ingests a PDF and returns the four fields in a machine-readable format (JSON or CSV is fine). I do not have a labelled dataset, however I can provide many example pdf files containing these shapes/profiles.
Deliverables
• Trained model (or inference script) that processes a PDF drawing and outputs Width, Length, Thickness and Material Grade with high accuracy
• Brief README explaining setup, dependencies and how to run the extraction locally
• Validation report that shows accuracy figures on a hold-out set of drawings I will provide
Acceptance Criteria
• ≥ 98 % dimensional accuracy measured against my test set
• Correct Material Grade identification on at least 95 % of samples
• Works on new unseen technical engineering PDFs without manual pre-editing
If this sounds straightforward to you, tell me how you plan to tackle dimension scaling, title-block detection and tolerance handling, and we can get started right away.
A more detailed project requirement is attached.
You are free to choose your preferred stack—Python with OpenCV, Tesseract, TensorFlow or any vision + NLP combination—so long as the final solution ingests a PDF and returns the four fields in a machine-readable format (JSON or CSV is fine). I do not have a labelled dataset, however I can provide many example pdf files containing these shapes/profiles.
Deliverables
• Trained model (or inference script) that processes a PDF drawing and outputs Width, Length, Thickness and Material Grade with high accuracy
• Brief README explaining setup, dependencies and how to run the extraction locally
• Validation report that shows accuracy figures on a hold-out set of drawings I will provide
Acceptance Criteria
• ≥ 98 % dimensional accuracy measured against my test set
• Correct Material Grade identification on at least 95 % of samples
• Works on new unseen technical engineering PDFs without manual pre-editing
If this sounds straightforward to you, tell me how you plan to tackle dimension scaling, title-block detection and tolerance handling, and we can get started right away.
A more detailed project requirement is attached.