Founder experience · Healthcare technology

Clinical NLP in production.

Clinical language is full of distinctions a convenient demo can miss: negation, timing, family history, terminology, and differences across languages. Kate worked on systems that have to handle those distinctions at operational scale.

Multilingual NLPExtractionEvaluationModel serving
The challenge

Make performance useful and repeatable.

A clinical NLP platform needs to extract structured information from text without losing the context that changes its meaning. Faster processing is valuable only when teams can see what it does to quality and release confidence.

Kate's role

Engineer and technical product lead.

In a healthcare technology team, Kate worked across multilingual extraction, model lifecycle, serving, evaluation, and release readiness. Her responsibilities included turning model behaviour into testable engineering and product decisions.

Concrete work

Named entity recognition and normalisation; negation, temporality, and family-history handling; relation extraction; weak labelling; and evaluation workflows for production changes.

What the work made possible

More systematic regression checking and performance improvements across components of the NLP stack. The point was to make changes inspectable before they reached a clinical workflow.

What this experience transfers

Evaluation design, model and rule tradeoffs, serving performance, and close collaboration between domain experts and engineers. These patterns matter wherever language systems must work beyond a curated demo.

Related challenge?

Let's inspect the real failure cases.

We can start with a sample, a workflow, and the decision your team needs to make.