Local Docling PDF Classification

Docling is already our preferred platform, but we have never installed or configured it on-premise. I need a clean setup on a single Windows workstation in our office and a repeatable, fully GDPR-compliant workflow that lets staff classify sensitive PDFs entirely offline.

The analysis we care about is data classification: each document must be checked for Personal Identifiable Information such as names, dates of birth and the frequency with which a customer interacts with the business. Nothing leaves the machine, and every temporary file or cache must be handled in line with GDPR data-minimisation rules.

What you will hand over
• Docling installed locally, with any required dependencies and OCR modules enabled for PDF ingestion
• A configuration (rules, taxonomy, or model) that reliably flags the above PII elements
• Step-by-step instructions or a small screencast showing non-technical staff how to run the workflow from import to export
• A brief compliance note explaining where files are stored, how logs are rotated and how to purge data

Acceptance criteria
1. All processing verified offline (no Internet callouts when the workstation is firewalled).
2. A sample test run classifies at least 95 % of known PII instances in a provided PDF set.
3. Documentation is clear enough for a colleague with no coding background to repeat the process.

If additional open-source libraries are useful, feel free to include them as long as everything stays on the local machine and licensing remains compatible with commercial use.