Marie Humbert-Droz

Machine Learning Scientist, Healthcare & Femtech Innovation

Helping healthcare teams build intelligent, reliable ML systems that scale — from clinical NLP to ovarian aging models.

Available for limited healthcare AI consulting.

About

I help healthcare teams turn complex clinical data into intelligent, trustworthy ML systems. My work sits at the intersection of clinical NLP, generative AI, and healthcare workflows — with a focus on interpretability, accuracy, and clinician trust.

I’m currently the Founding ML Engineer at Timeless Biotech, where I’m developing models and applications to help women understand and take control of ovarian aging, with the broader goal of extending healthspan and longevity. Previously, as Founding ML Scientist at Lighten, I built an LLM-powered abstraction platform to extract complex clinical variables from unstructured patient records at scale.

I also publish technical tutorials on Towards AI, helping teams and practitioners apply advanced ML to real-world healthcare problems.

Outside of work, I’m passionate about women’s health more broadly and coach strength training for women.

Expertise

Areas where I can help your team move faster, build smarter, and deliver real impact:

Large Language ModelsSurvival AnalysisTime-to-Event ModelingHealthcare Data SciencePredictive AnalyticsFemtech AI SolutionsStatistical ModelingDeep Learning

Featured Work

Case Study

Ovarian Aging Clock

Building a patent-pending machine learning platform to dramatically improve precision of time-to-menopause prediction. Developed the core ML library, API, database, and frontend application. The model reduces current uncertainty by nearly 10×, enabling more informed health decisions for women.

This framework lays the foundation for personalized menopause timeline prediction and longitudinal health planning.

Case Study

Healthcare LLM Platform

Developed a large language model system for clinical data abstraction of patient EHRs — often spanning thousands of pages per patient — to extract complex clinical information with high accuracy. Leveraged prompt engineering and agentic workflows to achieve F1 scores above 90%, enabling scalable, reliable clinical abstraction at scale.

This approach can be adapted to accelerate data abstraction for research, QA, or clinical workflows.