Anonymize Pro removes personal data from documents at scale. I was the primary technical owner of the platform, covering architecture, service integration, Docker-based deployment, and CI/CD delivery.
What I worked on
- Researched, designed, and built the core annotation engine, a Rust crate that detects personal data in Dutch government PDFs and classifies it under the Wet open overheid (Woo). It is the core engine powering the platform.
- Designed and built a fault-tolerant, tiered, fair-share job scheduler and dispatcher that shares a fixed pool of workers between paying tenants. Weighted service tiers give higher tiers more of the machine without letting any tenant starve the rest, and the design covers idempotent and atomic dispatch, cancellation, starvation prevention, and optional preemption.
- Built the asynchronous annotator worker that runs the annotation engine on documents arriving over Kafka.
- Built doc-layout-rs, a single contract shared PDF knowledge library used across services.
- Integrated DVC-based evaluation workflows (precision and recall) and added evaluation and integration tests to the CI pipelines for data quality assurance.