Clinical de-identification. Balance quality and speed.

PseudoPath detects identifying entities in clinical notes. Choose your NER model and use a trained router to send it only relevant lines, balancing detection quality with processing speed.

Explore measured performance with JavaScript enabled.

Start with PseudoPath

Choose a CamemBERT, DistilCamemBERT, or Tok2Vec preset, train it on your annotations, and return identifying entities as spaCy spans. Your application can use these spans to mask or replace the detected text.

Install with Python 3.11 or newer:

python -m pip install pseudopath

Follow the quickstart for optional Transformer dependencies and a first training run. Then explore:

• Models and configuration to choose a preset and tune its training.

• Training and routing to adapt a model and choose recall budgets.

• Custom models to integrate your own backend or checkpoint.

• Architecture for the workflow diagram, and reproducibility to record and compare experiments.