Principal Scientific Software Engineer | Kumar Lab

## Principal Scientific Software Engineer | Kumar Lab Apply:
Bar Harbor, Maine:
Full time:
Posted 17 Days Ago:
JR006164
** Job Summary
** The Kumar Lab studies the genetic and neurological basis of behavior with the goal of therapeutic and mechanistic discovery. We leverage machine learning and computer vision methods to model human diseases by transforming videos of mice into quantitative behavioral traits. Technology developed by our lab has been deployed in the JAX Envision System, a home cage monitoring platform. Envision streams continuous petabyte-scale video from animal housing into a cloud archive and operates on it in close to real time using multi-task algorithms.

Segmentation, pose estimation, and instance assignment are all used for action recognition, action localization tasks using supervised and unsupervised approaches to quantify complex animal behaviors representative of health and disease.

As Principal Scientific Software Engineer you will collaboratively shape the machine learning and computer vision engineering behind digital measures, harden pre-existing algorithms, and leverage robust MLOps principles. You will set technical direction, mentor trainees, and drive publications and model releases that come out of the work.

A successful candidate will be independently motivated, work collaboratively, and contribute meaningfully to the development of therapeutics for neurodevelopmental and neuropsychiatric disorders.
** The Challenges
** Mice are inherently difficult to study because they tend to avoid detection. As highly flexible, deformable animals, they are primarily active in low-light conditions and occupy small, confined spaces. These behaviors create significant challenges for computer vision tasks such as segmentation, pose estimation, instance tracking and identity tracking.
* ** Human Annotation is Expensive.
** Expert behaviorists’ time is limited, so creating a system that enables quick, high impact, scalable annotation is a must.
* ** Occlusion Complicates Behavior Annotation.
** Group-housed mice huddle and occlude each other.
* ** Models Must Generalize.
** Measures must perform across a diversity of genetic backgrounds, environments, coat colors, and sites, and across an archive too large to easily reprocess. Continual learning, edge-case mining, and efficient deployment are critical. You will be working with messy real-world data.
** What You'll Contribute
*** Set the technical direction for how the lab trains, evaluates, versions, and deploys computer vision models, including deployment to constrained edge hardware and fine-tuning of open-weight models for our downstream tasks.
* Collaborate with behavioral experts to shape the design, implementation, and deployment of behavior indices used to model human diseases in a production environment. Drive the publication and release of the resulting software.
* Own architecture decisions and their consequences: what gets built to last, what is deliberately temporary, and what we are choosing to pay for later. Set and enforce the standards — testing, documentation, review, reproducibility — that keep results repeatable after their author has left.
* Lead or co-lead evaluation of new methods and technologies, and how they would integrate with what we run today. Contribute to proposals and funding for the work you want to see done.
* Mentor lab members and contribute to technical leadership across the lab. Contribute directly to the technical growth of team members via proactive cross-team collaboration.
* Represent the lab as a senior technical voice inside JAX and externally, with collaborators, users of our released tools, and the broader research community.
** What You're Good At**
* ** Depth in machine learning and computer…