Member of Technical Staff, ML Training

Cognita Imaging • Palo Alto, CA

9h • Onsite

About The Position

Cognita's mission is to increase the world’s access to healthcare. Radiology is the first-line diagnostic specialty, facing a worsening global workforce shortage, and highly digitized, making it uniquely positioned for AI to have an enormous impact. Stage one of Cognita is focused on expanding access to radiology at scale. Our founding team met at Stanford, where they laid the groundwork for applying comprehensive AI to radiology. Building on that foundation, Cognita develops vision-language models that read radiology studies the way radiologists do - interpreting the full study in clinical context - and generate draft results that make radiologists more efficient and accurate. In partnership with Radiology Partners, Cognita’s models are trained and validated on one of the world’s largest real-world radiology datasets.

As a Member of Technical Staff focused on model training, you will be responsible for building Cognita’s large language models (LLMs) and vision-language models (VLMs) for radiology and architecting the training environment they run in. This role owns how models are trained: how data flows into training, how training jobs are structured, and how models improve through iteration and feedback.

Requirements

• Strong foundation in machine learning and deep learning.
• Experience training large models from scratch or architecting non-trivial training systems.
• Hands-on experience with LLMs and/or VLMs.
• Comfort owning ambiguous, high-impact systems end-to-end.

Nice To Haves

• Experience training large models in distributed environments.
• Familiarity with GPU-based training, memory optimization, or large-batch training.
• Experience with applied ML in production systems.
• Familiarity with medical imaging or clinical data (not required).

Responsibilities

• Architect and own the model training environment for large-scale models.
• Train LLMs / VLMs on radiology studies.
• Design training workflows that scale across large datasets and compute budgets.
• Define how data is cleaned, structured, versioned, and fed into training runs.
• Design model architectures, loss functions, and training objectives.
• Incorporate radiologist feedback and production signals into continuous model improvement.
• Investigate training instabilities, model regressions, and failure modes.
• Work closely with ML infrastructure engineers to ensure training systems are reliable, performant, and scalable.
• Partner with evaluation engineers to ensure training improvements result in measurable clinical gains.

What This Job Offers

• Job Type: Full-time
• Career Level: Senior
• Education Level: No Education Listed