Machine Learning Engineer, Clinical NLP

About the role

Meridian extracts structured data from clinical notes. The notes are messy, the schemas are strict, and a wrong extraction has consequences that are not measured in click-through rate.

What you will do

Build extraction pipelines from unstructured clinical text
Guarantee schema-valid output — downstream systems cannot handle malformed records
Work with clinicians on annotation guidelines and inter-rater agreement
Maintain evaluation as models and prompts change

What we look for

Named entity recognition or structured extraction in production
Experience with constrained generation or schema validation
Comfort working with domain experts who are not engineers
Care about correctness over throughput

What the assistant changed, and why

Change

Headline changed to name structured extraction directly

Why

You have done exactly this and called it "NLP pipelines", which does not match their language.

Change

Added the annotation guideline work with the agreement figure

Why

They name inter-rater agreement. You raised Cohen's kappa from 0.61 to 0.84 and had left it out entirely.

Change

Made the constrained decoding work explicit

Why

It was described as "JSON output". They ask for constrained generation, which is what you actually built.

Change

Removed the recommender project

Why

Nothing in this posting touches it, and space is better spent on extraction.

What matched

Structured extraction in production with a strict schema
Constrained decoding for schema-valid output
Annotation guideline work with a measured agreement improvement
Regular collaboration with non-engineer domain experts

What it would not stretch

No clinical or healthcare background. Not claimed. The posting does not require it, but expect it as a question.

Awaiting your approval

The draft is written and the form is filled in on Meridian Health’s own site. It sits here until you approve this one application. There is no approve-all, and nothing is submitted without you.