Senior Computer Vision Scientist - Digital Pathology

Contract | Mostly remote, Germany | January 2027 - February 2028 |

What if your next model helped decide which medicines are safe enough to reach patients?

Before any new medicine can be tested in humans, pathologists have to check tissue from preclinical studies for signs of toxicity. Slide by slide, every corner, even though most of it is perfectly healthy.

You could change that.

We're working with one of the world's largest research-driven pharmaceutical companies on an AI tool that flags the tissue regions pathologists actually need to look at. Early estimates suggest it could make the whole process around 40% faster. Their leadership calls it "a no-brainer."

What you’ll be doing

You'll take a published abnormality detection prototype, currently trained on a single organ, and turn it into something much bigger. You'll rebuild it on an in-house histopathology foundation model, a vision transformer trained with JEPA on more than 100,000 whole slide images. Then you'll scale it across multiple organs and a far larger dataset, prove it works through rigorous evaluation, and make it production-ready.

This one won't end up in a drawer. Your model will go into the software the toxicology pathologists use every day, supporting safety assessment across the company's late preclinical pipeline.

Why you’ll want this one

You'll work with data almost nobody else has. Most pathology AI focuses on human oncology. You'll be working with proprietary preclinical toxicology data across multiple organs, at a serious scale.

You can ship and publish

If the results are strong, the team plans to publish in a good journal, and if you've contributed, your name goes on the paper. It depends on the results, but how often does an industry contract give you both?

You’ll take it all the way to the regulator

You'll help lay the groundwork for GLP validation. Few ML specialists have that experience, and it will make your CV stand out as AI moves deeper into regulated healthcare.

You won’t start from scratch

There's a working foundation model, a published prototype, and written backing from senior leadership.

You’ll own it

The downstream model is yours to design, train, evaluate and improve. You'll work in a small, senior team alongside the foundation model developers, a computational pathologist who will plan annotation and priorities with you, and the engineers integrating your work.

You’ll work from home.

The role is mostly remote, with the occasional in-person meetup.

You’ll have stability. It's a 14-month contract with a good chance of continuing.

Is this you?

You have deep, hands-on computer vision expertise and real digital pathology experience. Whole slide images, gigapixel sizes and pyramidal formats don't faze you. You know your way around vision transformers, you enjoy designing proper experiments, and you can explain your work to pathologists without losing them in the math's.

You'll need the right to work in Germany, or in the EU with experience working for German companies.

It's a bonus if you have a PhD (a strong master's with the right experience works too), experience with JEPA or other self-supervised methods, experience building on foundation models, exposure to GLP or validation processes, or a publication record in computer vision or computational pathology.

#J-18808-Ljbffr

Data Science Talent Ltd