Qwen 3 and Paperless-ngx: document scanner setup

This walkthrough covers a document scanner setup that turns paper into searchable PDFs with metadata. The speaker first adapts a scanner requiring 24V to an existing 12V DC supply. He places a boost converter upstream to reduce cable losses and uses a step-down converter for another 12V load. His measurements suggest a small efficiency benefit for his installation.

The software workflow starts with image-only PDFs on a network file server. The speaker discusses Tesseract and OCRMyPDF as open source OCR options, but uses an unnamed commercial OCR appliance because it produced better results in his tests. It watches an input folder and writes PDFs with a text layer behind the scanned image. He also explains PDF/A for archival storage.

A server script starts the OCR virtual machine when the scanner responds to ping or documents appear in its folder. It shuts the VM down after the scanner turns off and the input folder remains empty for a while.

For metadata, he runs Qwen 3 32B as a local LLM on the CPU rather than the GPU. Processing takes several minutes; he estimates that suggestions are roughly correct in 80 to 90% of cases and still reviews them. He tests Paperless-ngx alongside his own document management system, using its API to import documents and metadata. The workflow functions, but his long-term choice of document manager remains unsettled.