Problem
A large European manufacturer producing custom-made products struggled with a slow, tedious quotation process for incoming orders and RFQs. The root cause was the wide variety of input formats, ranging from clean PDFs generated by specialized software to photos of handwritten notes and sketches.
Understanding, interpreting, and entering order parameters into the system was highly time-consuming and required significant experience. As a result, throughput and delivery time depended heavily on the availability of key employees.
Challenge
The biggest challenge was achieving high accuracy when extracting order parameters, which could number in the hundreds, while keeping the solution stable across varying input quality.
Solution
We built a solution based on multimodal AI algorithms. The system recognizes text, including handwritten text in multiple languages, analyzes technical drawings by classifying their type, and automatically extracts parameters.
The system combines several components, including direct integrations with multimodal model APIs (OpenAI), advanced prompt engineering, and fine-tuned computer vision models for detecting details in the inputs and classifying drawing types.
The outputs are processed and delivered as JSON to the order-intake system, integrating seamlessly with the company’s existing infrastructure.
Result
The solution significantly reduces order-handling time, improves data-entry accuracy, and relieves employees so they can focus on higher-value work.
